U.S. flag

An official website of the United States government

The .gov means it's official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you're on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

  • Publications
  • Account settings
  • Browse Titles

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Iliffe S, Wilcock J, Drennan V, et al. Changing practice in dementia care in the community: developing and testing evidence-based interventions, from timely diagnosis to end of life (EVIDEM). Southampton (UK): NIHR Journals Library; 2015 Apr. (Programme Grants for Applied Research, No. 3.3.)

Cover of Changing practice in dementia care in the community: developing and testing evidence-based interventions, from timely diagnosis to end of life (EVIDEM)

Changing practice in dementia care in the community: developing and testing evidence-based interventions, from timely diagnosis to end of life (EVIDEM).

Appendix 65 chapter 5 : five main stages in framework analysis.

  • Familiarisation

This early stage is for the researchers to get familiarised with the data and sensitised to early themes. It encourages the research to see the individual differences inherent in transcripts that can sometimes get lost when coding begins. The process of sensitisation to these individual differences also enables the researcher to better identify within- and between-participant differences. In a few cases, the researcher felt the need to revert to the original recorded interview to get a better feel for the data. Any emergent early impressions were noted, relating to reactions to transcript/participant, how they felt as they listened to the participant, and any specifics that they wanted to remember for later. These were jotted on one side of the paper transcript.

  • Identifying a thematic framework

This stage of framework analysis is commonly referred to as ‘coding’ in other qualitative methodologies. This principally involves identifying key themes, issues or discussion points embedded in the transcript. These are delineated and assigned a ‘code’ or a name that best captures the essence of the theme or issue identified. In framework analysis, ‘a priori issues’ questions can form the basis of the themes – we therefore used the interview topic guide as a starting point for creating overarching categories and any emergent themes from transcripts were coded as responses to each question. As Pope and colleagues describe, 262 the outcome of this stage is a ‘detailed index’ of the data into ‘manageable chunks for subsequent retrieval and exploration’, which is what we achieved.

  • We were keen to ascertain the training on the MCA practitioners had been offered. We therefore labelled ‘Training’ as one of our a priori categories and coded (1) ‘no training’, (2) ‘1-day training course’, (3) ‘1+ day training programme’, and (4) ‘MCA training as part of other activities’. We were able to identify and make inferences regarding the amount of training one participant group had had compared with another.

Indexing refers to the process of numerically annotating transcripts in order to identify consistencies, which then go on to develop the coding framework. However, we followed this process slightly differently. All of the word codes (as opposed to numerical) that had been generated during stage 2 were listed on a separate sheet of paper. We then grouped together all codes that shared commonalities or consistencies. These clusters were given an appropriate name.

  • One of our interview questions referred to ‘Practice experiences’ – we named this as an overarching category, incorporating (1) ‘intrinsic impact’, (2) ‘community role of X’, (3) ‘role of safeguarding’, and (4) ’change and extension to job role’ as separate codes under it. 269
  • This also helped us to identify and be mindful of individual participant differences. For example, while community nurses saw safeguarding as less of a priority in their jobs during time 1 interviews 258 SACs talked about the significant impact of the MCA on their practice experiences. 264

Framework analysis describes this stage as a process of rearranging the data and thematic framework to create order, not dissimilar to the iterative principle of grounded theory. 449 We applied this stage as a principle for synthesising and developing our final coding framework through a process of abstraction, in order to derive all the detail from the data and ensure that we were coding elements that may have been missed with simply an a priori approach.

  • The question asking all participants about their personal experiences of caregiving (family caring) was asked as a final question to round off the interview. Family caring experiences were commonly reported and participants offered examples of how these had contributed to their professional understanding of dementia and caring. In relation to this, five main codes were developed: (1) ‘informing the professional role’, (2) ‘insight into services’, (3) ‘professional influences on caring’, (4) ‘planning’, and (5) ‘no apparent effect of MCA’.
  • Mapping and interpretation

Mapping and interpreting essentially are ways of representing pictorially or graphically all of the themes and investigating how each of the themes relates to each other. This detailed exploration of the iteratively developed and revised thematic framework enabled us to gain a clearer understanding and explanation of the ‘bigger picture’. For instance, we examined the association between three categories: (1) ‘training’, (2) ‘(self-reported) knowledge of the MCA’, and (3) ‘self-reported confidence’; and we were able to identify and offer explanations for individual discrepancies between these three categories.

  • In Manthorpe et al. 268 we describe how some participants, with no training and limited self-reported knowledge, reported high levels of confidence in abiding by the Act. We questioned the reliability of some self-reported knowledge about the MCA, and have considered the adequacy of training now the Act is no longer ‘new’ and the potential for mandatory training for all staff.

Included under terms of UK Non-commercial Government License .

  • Cite this Page Iliffe S, Wilcock J, Drennan V, et al. Changing practice in dementia care in the community: developing and testing evidence-based interventions, from timely diagnosis to end of life (EVIDEM). Southampton (UK): NIHR Journals Library; 2015 Apr. (Programme Grants for Applied Research, No. 3.3.) Appendix 65, Chapter 5: Five main stages in framework analysis.
  • PDF version of this title (7.0M)

In this Page

Other titles in this collection.

  • Programme Grants for Applied Research

Recent Activity

  • Chapter 5: Five main stages in framework analysis - Changing practice in dementi... Chapter 5: Five main stages in framework analysis - Changing practice in dementia care in the community: developing and testing evidence-based interventions, from timely diagnosis to end of life (EVIDEM)

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

Connect with NLM

National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894

Web Policies FOIA HHS Vulnerability Disclosure

Help Accessibility Careers

statistics

  • Privacy Policy
  • SignUp/Login

Research Method

Home » Framework Analysis – Method, Types and Examples

Framework Analysis – Method, Types and Examples

Table of Contents

Framework Analysis

Framework Analysis

Definition:

Framework Analysis is a qualitative research method that involves organizing and analyzing data using a predefined analytical framework. The analytical framework is a set of predetermined themes or categories that are derived from the research questions or objectives. The framework provides a structured approach to data analysis and can help to identify patterns, themes, and relationships in the data.

Steps in Framework Analysis

Here are the general steps in Framework Analysis:

Familiarization

Get familiar with the data by reading and re-reading it. This step helps you to become immersed in the data and to get a sense of its content, structure, and scope.

Identify a Coding Framework

Identify a coding framework or set of themes that will be used to analyze the data. These themes can be derived from existing literature or theories or developed based on the data itself.

Code the data by applying the coding framework to the data. This involves breaking down the data into smaller units and assigning each unit to a particular theme or category.

Chart or summarize the data by creating tables or matrices that display the distribution and frequency of each theme or category across the data set.

Mapping and interpretation

Analyze the data by examining the relationship between different themes or categories, and by exploring the implications and meanings of the findings in relation to the research question.

Verification

Verify the accuracy and validity of the findings by checking them against the original data, comparing them with other sources of data, and seeking feedback from others.

Report the findings by presenting them in a clear, concise, and organized manner. This involves summarizing the key themes, presenting supporting evidence, and providing interpretations and recommendations based on the findings.

Framework Analysis Conducting Guide

Here is a step-by-step guide to conducting framework analysis:

  • Define the research question: The first step in conducting framework analysis is to clearly define the research question or objective that you want to investigate.
  • Develop the analytical framework: Develop a coding framework or a set of predetermined themes or categories that are relevant to the research question. These themes or categories can be derived from existing literature or theories, or they can be developed based on the data collected.
  • Data collection: Collect the data using a suitable method such as interviews, focus groups, surveys or observation.
  • Familiarization: Transcribe and familiarize yourself with the data. Read through the data several times and take notes to identify any patterns, themes or issues that are emerging.
  • Coding : Code the data by identifying key themes or categories and assigning each piece of information to a specific theme or category.
  • Charting: Create charts or tables that display the frequency and distribution of each theme or category. This helps to summarize the data and identify patterns.
  • Mapping and interpretation: Analyze the data to identify patterns, relationships, and themes. Interpret the findings in light of the research objectives and provide explanations for any significant patterns or themes that have emerged.
  • Validation : Validate the findings by sharing them with others and seeking feedback. This can help to ensure that the findings are robust and reliable.
  • Report writing: Write a report that summarizes the findings, includes quotes or examples from the data to support the findings and provides recommendations for future research.

Applications of Framework Analysis

Framework Analysis has a wide range of applications in research, including:

  • Policy analysis: Framework Analysis can be used to analyze policies and policy documents to identify key themes, patterns, and underlying assumptions.
  • Social science research: Framework Analysis is commonly used in social science research to analyze qualitative data from interviews, focus groups, and other sources.
  • Health research: Framework Analysis can be used to analyze qualitative data from health research studies, such as patient and provider perspectives, to identify themes and patterns.
  • Environmental research : Framework Analysis can be used to analyze qualitative data from environmental research studies to identify themes and patterns related to environmental attitudes, behaviors, and practices.
  • Education research: Framework Analysis can be used to analyze qualitative data from educational research studies to identify themes and patterns related to teaching practices, student learning, and educational policies.
  • Market research: Framework Analysis can be used to analyze qualitative data from market research studies to identify themes and patterns related to consumer attitudes, behaviors, and preferences.

Examples of Framework Analysis

Here are some examples of Framework Analysis in various research contexts:

  • Health Research: A study on the experiences of cancer survivors might use Framework Analysis to identify themes related to the psychological, social, and physical aspects of survivorship. Themes might include coping strategies, social support, and health outcomes.
  • Education Research: A study on the impact of a new teaching approach might use Framework Analysis to identify themes related to the implementation of the approach, the effectiveness of the approach, and barriers to its implementation. Themes might include teacher attitudes, student engagement, and logistical challenges.
  • Environmental Research : A study on the factors that influence pro-environmental behaviors might use Framework Analysis to identify themes related to environmental attitudes, behaviors, and practices. Themes might include social norms, personal values, and perceived barriers to behavior change.
  • Policy Analysis: A study on the implementation of a new policy might use Framework Analysis to identify themes related to policy development, implementation, and outcomes. Themes might include stakeholder perspectives, organizational structures, and policy effectiveness.
  • Social Science Research: A study on the experiences of immigrant families might use Framework Analysis to identify themes related to the challenges and opportunities faced by immigrant families in their new country. Themes might include language barriers, cultural differences, and social support.

When to use Framework Analysis

Framework Analysis is a useful method for analyzing qualitative data when the research questions require an in-depth exploration of a particular phenomenon, concept, or experience. It is particularly useful when:

  • The research involves multiple sources of qualitative data, such as interviews, focus groups, or documents, that need to be analyzed and compared.
  • The research questions require a systematic and structured approach to data analysis that enables the identification of patterns, themes, and relationships in the data.
  • The research involves a large and complex dataset that requires a method for organizing and synthesizing the data in a meaningful way.
  • The research aims to generate new insights and understandings from the data, rather than testing pre-existing hypotheses or theories.
  • The research requires a method that is transparent, replicable, and verifiable, as Framework Analysis provides a clear framework for data analysis and reporting.

Purpose of Framework Analysis

The purpose of Framework Analysis is to systematically organize and analyze qualitative data in a structured and transparent manner. The method is designed to identify patterns, themes, and relationships in the data that are relevant to the research question or objective. By using a rigorous and transparent approach to data analysis, Framework Analysis enables researchers to generate new insights and understandings from the data, and to provide a clear and structured presentation of the findings.

The method is particularly useful for analyzing large and complex qualitative datasets that require a method for organizing and synthesizing the data in a meaningful way. It can be used to explore a wide range of research questions and objectives across various fields, including health research, social science research, education research, policy analysis, and environmental research, among others.

Overall, the purpose of Framework Analysis is to provide a systematic and transparent method for analyzing qualitative data that enables researchers to generate new insights and understandings from the data in a rigorous and structured manner.

Characteristics of Framework Analysis

Some Characteristics of Framework Analysis are:

  • Systematic and Structured Approach: Framework Analysis provides a systematic and structured approach to data analysis that involves a series of steps that are followed in a predetermined order.
  • Transparency and Replicability: Framework Analysis emphasizes transparency and replicability, as it involves a clearly defined process for data analysis that can be applied consistently across different datasets and research questions.
  • Flexibility : Framework Analysis is flexible and adaptable to a wide range of research contexts and objectives, as it can be used to analyze qualitative data from various sources and to explore different research questions.
  • In-depth Exploration of the Data: Framework Analysis enables an in-depth exploration of the data, as it involves a thorough and detailed analysis of the data to identify patterns, themes, and relationships.
  • Applicable to Large and Complex Datasets: Framework Analysis is particularly useful for analyzing large and complex qualitative datasets, as it provides a method for organizing and synthesizing the data in a meaningful way.
  • Data-Driven: Framework Analysis is data-driven, as it focuses on the analysis and interpretation of the data rather than on pre-existing hypotheses or theories.
  • Emphasis on Contextual Understanding : Framework Analysis emphasizes contextual understanding, as it involves a detailed examination of the data to identify the social, cultural, and environmental factors that may influence the phenomena under investigation.

Advantages of Framework Analysis

Some Advantages of Framework Analysis are as follows:

  • Transparency : Framework Analysis provides a clear and structured approach to data analysis, which makes the process transparent and easy to follow. This ensures that the findings can be easily replicated or verified by other researchers.
  • Thorough Analysis : Framework Analysis enables a thorough and detailed analysis of the data, which allows for the identification of patterns, themes, and relationships that may not be apparent through other methods.
  • Contextual Understanding: Framework Analysis emphasizes the importance of understanding the context in which the data was collected, which enables a more nuanced interpretation of the findings.
  • Collaborative Analysis: Framework Analysis can be used as a collaborative method for data analysis, as it allows multiple researchers to work together to analyze the data and develop a shared understanding of the findings.
  • Efficient and Time-saving: Framework Analysis can be an efficient and time-saving method for analyzing qualitative data, as it provides a structured and organized approach to data analysis that can help researchers manage and synthesize large datasets.
  • Comprehensive Reporting: Framework Analysis can help ensure that the research findings are comprehensive and well-reported, as the method provides a clear framework for presenting the results.

Limitations of Framework Analysis

Some Limitations of Framework Analysis are as follows:

  • Subjectivity : Framework Analysis relies on the interpretation of the researchers, which can introduce subjectivity into the analysis process.
  • Time-consuming : Framework Analysis can be a time-consuming method for data analysis, particularly when working with large and complex datasets.
  • Limited ability to generate new theory : Framework Analysis is a deductive approach that relies on pre-existing theories and concepts to guide the analysis, which may limit the ability to generate new theoretical insights.
  • Risk of oversimplification: The structured approach of Framework Analysis can lead to oversimplification of the data, as complex issues may be reduced to predefined categories or themes.
  • Limited ability to capture the complexity of the data : The predefined categories or themes used in Framework Analysis may not be able to capture the full complexity of the data, particularly when dealing with nuanced or context-specific phenomena.
  • Limited use with non-textual data : Framework Analysis is primarily designed for analyzing qualitative textual data and may not be as effective for analyzing non-textual data such as images, videos, or audio recordings.

About the author

' src=

Muhammad Hassan

Researcher, Academic Writer, Web developer

You may also like

Probability Histogram

Probability Histogram – Definition, Examples and...

Substantive Framework

Substantive Framework – Types, Methods and...

Factor Analysis

Factor Analysis – Steps, Methods and Examples

Graphical Methods

Graphical Methods – Types, Examples and Guide

Critical Analysis

Critical Analysis – Types, Examples and Writing...

Grounded Theory

Grounded Theory – Methods, Examples and Guide

Book cover

Revealing Media Bias in News Articles pp 109–140 Cite as

Frame Analysis

  • Felix Hamborg 2  
  • Open Access
  • First Online: 06 October 2022

2355 Accesses

This chapter details the last component of person-oriented framing analysis: frame analysis. The component aims to classify how persons are portrayed in news articles. The chapter introduces and discusses two approaches for this task. First, it briefly presents an exploratory approach that aims to classify fine-grained categories of how persons are portrayed. Afterward, the chapter introduces the first method for target-dependent sentiment classification in the domain of news articles. The dataset and method enable sentiment classification in a domain that could not reliably be analyzed earlier. Lastly, the chapter argues for using the latter approach in the frame analysis component, in particular because of its high classification performance.

Download chapter PDF

5.1 Introduction

This chapter describes the frame analysis component, which aims to identify how persons are portrayed in the given news articles, both at the article and sentence levels. This task is of difficulty for various reasons, such as that news articles rather implicitly or indirectly frame persons, for example, by describing actions performed by a person. Consequently, a high level of interpretation is necessary to identify how news articles portray persons. Because of this and other issues highlighted later in this chapter, prior approaches to analyze frames or derivatives yield inconclusive or superficial results or require high manual effort, e.g., to create large annotated datasets.

Frame analysis is the second and last analysis component in person-oriented framing analysis (PFA). The input to frame analysis is a set of news articles reporting on the same policy event, including the persons involved in the event and their mentions across all articles. The output of the frame analysis component should be information concerning how each person is portrayed and groups of articles similarly framing the persons involved in the event.

An approach in principle suitable for the frame analysis component is identifying political frames as defined by Entman [ 79 ]. Doing so would approximate the content analysis, particularly the frame analysis, as conducted in social science research on media bias most closely. However, as pointed out in Sect. 3.3.2 , taking such an approach would result in infeasibly high annotation cost. Further, political frames are defined for a specific topic or analysis question [ 79 ], whereas PFA is meant to analyze bias in news coverage on any policy event. Thus, identifying political frames seems out of the thesis’s scope and also methodologically infeasible currently. We revisit this design decision later in the chapter and also in our discussion of the thesis’s limitations and future work ideas in Sect. 7.3.2 .

In this chapter, we explore two conceptually different approaches to determine how individual persons are portrayed. The first approach more closely resembles how researchers in the social sciences analyze framing. It aims to identify categories representing topic-independent framing effects, which we call frame properties , such as whether a person is portrayed as being competent, wise, powerful, or trustworthy (or not). The second approach follows a more pragmatic route to the task of the frame analysis component, which we devise to address fundamental issues of the first approach, such as high annotation cost, high annotation difficulty, and low classification performance. Both approaches have in common that they do not analyze frames, which would be the standard procedure in the social sciences, but instead categorized effects of framing. We focus on framing effects since frames as analyzed in social science research on media bias are topic-specific. In contrast, our approach is meant to analyze news coverage on any policy issue (see also Sects. 1.3 and 3.3.2 ).

Note that the frame analysis component consists of an additional second task, i.e., frame clustering (Fig. 3.1 ). Once the frame analysis has identified how each person is portrayed at both the article and sentence levels, frame clustering aims to find groups of articles that frame the persons similarly. For frame clustering, we use a simple technique that we will describe when introducing our prototype system (Chap. 6 ).

5.2 Exploring Person-Targeting Framing(-Effects) in News Articles

The section presents the results of a research direction we pursued initially for the frame analysis component. We explore a simple approach that aims to identify so-called frame properties, which are fine-grained constituents of how a person is portrayed at the sentence level, e.g., whether a person is shown as competent or powerful. Frame properties resemble framing effects in social science research. We propose frame properties as pre-defined categorical characteristics that might be attributed to a target person due to one or more frames. For example, in a sentence that frames immigrants as intruders that might harm a country’s culture and economy (rather than victims that need protection, cf. [ 369 ]), respective frame properties of the mentioned immigrants could be “dangerous” and “aggressive.”

The remainder of this section is structured as follows. Section 5.2.1 briefly summarizes prior work on automated frame analysis. We then present our exploratory approach for frame property identification in Sect. 5.2.2 . Afterward, we discuss the results of an exploratory, qualitative analysis in Sect. 5.2.3 . Section 5.2.4 highlights the shortcomings and difficulties of this approach and discusses how to address or avoid them. Specifically, our approach achieved in the evaluation only mixed results, but we can use the identified issues to derive our main approach for the frame analysis component, which is then described in Sect. 5.3 . Lastly, Sect. 5.2.5 provides a brief summary of our exploratory research on frame property identification.

The dataset used for the approach and its codebook are available at

https://github.com/fhamborg/NewsWCL50 .

5.2.1 Related Work

This section briefly summarizes key findings of our literature review concerning the analysis of political framing. An in-depth discussion of the following and other related approaches can be found in Chap. 2 .

To analyze how persons (or other semantic concepts) are portrayed, i.e., framed, researchers from the social sciences primarily employ manual content analyses and frame analyses (see Sect. 2.3.4 ). In content analyses focusing on the person-targeting bias forms (Sect. 3.3.2 ), social scientists typically analyze how news articles frame individual persons or groups of persons. For example, whether there is a systematic tendency in coverage to portray immigrants in a certain way, such as being aggressive or helpless [ 263 ]. Observing these tendencies may then yield specific frames, such as the “intruder” or “victim” frames mentioned in the beginning of Sect. 5.2 . Other analyses not concerned with persons but, e.g., topics, may focus on whether news outlets use emotional or factual language when reporting on a specific topic or which topical aspects of an issue are highlighted in coverage [ 274 ].

In sum, in our literature review on identifying media bias, we find that no automated system focuses on the analysis of person-targeting bias forms at the sentence level (see Sects. 2.3.2 – 2.3.4 ). However, two prior works are of special interest. First, the research conducted for the creation of the media frame corpus (MFC) aims to directly represent political frames [ 80 ] as established in the social sciences [ 45 ]. In contrast to political frames, MFC’s “frame types” are topic independent and thus are in principle highly relevant for our task. However, from a conceptual perspective, the MFC’s frame types are independent of any target, i.e., they holistically describe the content or “frame” of a news article or a sentence within. Moreover, this approach suffers from high annotation cost and low inter-coder reliability (ICR) [ 45 ]. As a consequence, classifiers trained on the MFC yield low classification accuracy on the sentence level [ 46 ].

Second, a recent approach aims to automatically extract so-called microframes from a set of text documents, e.g., news articles [ 193 ]. Given a set of text documents, the respective microframes are defined as semantic axes that are over-represented in individual documents. Each such semantic axis is a bi-polar adjective pair as used in semantic differential scales established in psychology [ 145 ]. Since the microframes are extracted for a given set of documents, they are topic-dependent or, more specifically, dataset-dependent. For example, in a topic and corresponding documents on immigration, the adjective pair of one microframe could be “illegal”-“legal.” After the extraction of these microframes for a given dataset, users then review the microframes and select a subset of them to be used for further analysis. The approach yields qualitative microframes that resemble closely our frame properties but are, in contrast to them, dataset-specific.

Most of the other reviewed approaches that are only partially related to our task use quantitative methodology, and their results are mostly superficial, especially when compared to the results of manual content analyses. For example, one approach investigates the frequency of affective words close to user-defined words [ 116 ], e.g., names of politicians. Another approach aims to find bias words by employing IDF [ 211 ].

Another field that is relevant for the task of determining how persons are portrayed is sentiment analysis and more specifically target-dependent sentiment classification [ 212 ]. However, researchers have questioned whether the one-dimensional polarity scale of sentiment classification suffices to capture the actual fine-grained effects of framing (see Sect. 2.3.4 ). We will investigate sentiment classification and this question in Sect. 5.3 .

5.2.2 Method

Given a person mention and its context, the objective of our method is to determine which frame properties the context as well as the mention yield on the person. Specifically, our simple method looks for words that express one or more frame properties on the person mention. Afterward, we aggregate for each target person frame properties from all its mentions in the given news article from sentence level to article level. Footnote 1

The idea of our approach is to extend the one-dimensional polarity scale, i.e., positive, neutral, and negative, established in traditional sentiment classification with further classes, i.e., fine-grained properties, such as competent, powerful, and antonyms thereof. We call these fine-grained properties frame properties . Conceptually, some frame properties can be subsumed using sentiment polarity, but they also extend the characteristics that can be represented using polarity. For example, while affection and refusal can be represented as more specific forms of positive and negative portrayal, respectively, other frame properties cannot be projected into the positive-negative scale. Such frame properties include, for example, (being portrayed as an) aggressor versus a victim, where the victim has neither positive nor negative sentiment polarity (the aggressor, though, has clearly negative polarity). This way, our approach seeks to overcome the shortcomings of the one-dimensional polarity scale used in sentiment classification (see Sect. 2.3.4 ).

Using a series of inductive and deductive small-scale content analyses, we devised in total 29 frame properties, of which 13 are bi-polar pairs and 3 have no antonym. Specifically, to derive the final set of frame properties, we initially asked coders to annotate any phrase that they felt was influencing their assessment of a person or other semantic concepts mentioned and also to state which perception, judgment, or feeling the phrase caused in them. We then used these initial open statements to derive a set of frame properties, which represented how the annotators felt a target was portrayed. Table 5.1 shows the final set of frame properties.

We refined these initial frame properties in an interactive process following best practices from the social sciences until three goals or factors were achieved or maximized. First, an acceptable inter-coder reliability was reached. Second, the set of frame properties covers a broad spectrum of person characteristics highlighted by “any” news coverage while, third, still being as specific as possible. The second and third goals aim to achieve a balance between being topic-independent and generic while preferring specific categories, such as “competent,” over general categories, such as “positive.” We achieved these goals after conducting six test iterations consisting of reviewing previous annotation results, refining the codebook and frame properties, and discussion of the changes with the annotators. We reached an acceptable inter-coder reliability of 0.65 (calculated with mean pairwise average agreement). The comparably low inter-coder reliability indicates the complexity and difficulty of the task (cf. [ 45 ]). Further information on the training prior to the main annotation is described in Sect. 4.3.2.2 .

Finally, we conducted a deductive content analysis on 50 news articles reporting on policy issues using the codebook created during the training annotations. In the main annotation, 5926 mentions of persons and other target concepts were annotated. Further, 2730 phrases that each induce at least one frame property were annotated. For each frame property, additionally, the corresponding target concept had to be assigned. For example, in “Russia seizes Ukrainian naval ships,” “Russia” would be annotated as a target concept of type “country,” and “seizes” as a frame property with type “Aggressor” that targets “Russia.” Each mention of a target concept in a text segment can be targeted by multiple frame property phrases. Further information on the training prior to the main annotation is described in Sect. 4.3.2.3 .

After the annotation and in a one-time process, we manually defined a set of seed words for each of the frame properties S k  ∈  S . For each frame property S k , we gathered seed words by carefully selecting common synonyms from a dictionary [ 233 ], e.g., for the frame property “affection,” we selected the seed words: attachment, devotion, fondness, love, and passion.

For each news article passed to the frame analysis component, our method performs the following procedure. First, to identify frame property words, the method iterates all words in the given news article and determines for each word its semantic similarity to each of the frame properties. Specifically, we calculate the cosine similarity of the current word w and each seed word s  ∈  S k of the current frame property S k in a word embedding space [ 330 ]. We define the semantic similarity

We assign to a word w any frame property S k , where \(\mathrm {sim}\left ( w,S_{k} \right ) > t_{p} = 0.4\) . At the end of this procedure, each word has a set of weighted frame properties. The weight of a frame property on a word is defined by \(\mathrm {sim}\left ( w,S_{k} \right )\) .

Second, for each target person c i , we aggregate frame properties S k  ∈  S from all its modifiers w j of c i found by dependency parsing [ 6 ]. We use manually devised rules to handle the different types of relations between head c i and modifier w j , e.g., to assign the frame properties of an attribute (modifier) to its noun (target person mention) or a predeterminer (modifier) to its head (target person mention).

Given a news article and a set of persons or other semantic concepts with one or more mentions, the output of the proposed method is as follows. For each mention, the method determines a set of weighted frame properties, yielded by the sentence of the mention. Further, for each semantic concept, the method returns a set of weighted frame properties by aggregating them from mention level to article level.

5.2.3 Exploratory Evaluation

We discuss the usability of this exploratory approach as to determining frame properties in a set of news articles reporting on the same event in two use cases. Due to the low inter-coder reliability of the frame property annotations (see Sect. 5.2.2 ), we expected low classification performance of our approach. Thus, we did not conduct a quantitative evaluation but instead qualitatively investigated the approach to derive future research ideas [ 46 ]. In contrast to the research objective of this thesis, i.e., identifying and communicating biases targeting persons, in this investigation, we also considered semantic concepts of the type groups of persons. This allows us to better demonstrate and discuss the results of the method.

In the first use case, we investigated the frame properties of persons and other semantic concepts in an event, where the DNC, a part of the Democratic Party in the USA, sued Russia and associates of Trump’s presidential campaign in 2018 (see event #3 in Table 4.9 ). Table 5.2 shows exemplary frame properties of the three main actors involved in the event, Donald Trump, the Democratic Party, and the Russian Federation, each being a different concept type (shown in parentheses in Table 5.2 ). The first column shows each candidate’s representative phrase (see Sect. 4.3.3.1 ). The linearly normalized scores s ( c , a , f ) in the three exemplary frame property columns represent how strongly each article a (row) portrays a frame property f regarding a candidate c : s  = 1 or − 1 indicates the maximum presence of the property or its antonym, respectively. A value of 0 indicates the absence of the property or equal presence of the property and its antonym.

Left-wing outlets (LL and L) more strongly ascribe the property “aggressor” to Trump, e.g., s (Trump, LL, aggressor) = 1, than right-wing outlets do, for example, s (Trump, R, aggressor) = 0.34. This is conformal with the findings of manual analyses of news coverage of left- versus right-wing outlets regarding Republicans [ 71 , 116 , 117 ]. The Democratic Party is portrayed in all outlets as rather aggressive ( s  = [0.91, 1]), which can be expected due to the nature of the event, since the DNC sued various political actors.

The difficulty of frame property classification is visible in other frame properties that yielded inconclusive trends, such as “reason.” We found that an increased level of abstractness is the main cause for lower frame identification performance (cf. [ 45 , 211 , 276 ]). For example, in the content analysis (see Sect. 4.3.2 ), we noticed that “reason” was often not induced by single words but rather more abstractly through actions that were assessed as reasonable by the human coders.

In the second use case, we investigated frame properties in an event where special counsel Mueller provided a list of questions to Trump in 2018 (see event #8 in Table 4.9 ). Table 5.3 shows selected frame properties of the two main actors involved in the event: Trump and Mueller. Since both are individual persons, their semantic concept type is “Actor.” The results of our method indicated that the reviewed left-wing outlets ascribe rather positive frame properties to Mueller, e.g., s (Mueller, LL, confidence) = 1, than right-wing outlets do, s (…, RR, …) = 0. For Trump, we identified the opposite, e.g., s (Trump, LL, trustworthiness) = −0.19 and s (…, RR, …) = 1. More strongly, left-wing news outlets even ascribe non-trustworthiness to Trump, e.g., s (Trump, LL, trustworthiness) = −0.93. Besides these expected patterns, other frame properties again showed inconclusive trends, such as power.

Due to the difficulty of automatically estimating frames (see Sect. 2.5 ), the identification of frame properties ascribed to persons and other semantic concepts did not always yield clear or expected patterns. We found this is especially true for abstract or implicitly ascribed frame properties. For example, we could not find clear patterns for the frame properties “reason” in the first use case (Table 5.2 ) and “power” in the second use case (Table 5.3 ), which is mainly due to the abstractness used to portray a person as being powerful or reasonable.

5.2.4 Future Work

In our exploratory evaluation, we found that our basic approach yielded trends concerning frame properties that are in part as expected and in part inconclusive. We think there are two main causes for the inconclusive results. First, the simplicity of the approach is one key reason. The second potential cause is the general difficulty of annotating or determining frames and, respectively, in our case the frames’ effects [ 45 ].

To address the first issue, we propose to improve the approach using more sophisticated techniques, such as deep learning and recent language models. Fundamentally, our current approach is word-based and may often fail to catch the “meaning between the lines” (see Sect. 2.2.5 ). This is in contradiction to the substantial character of frames [ 80 ], which typically requires a higher degree of interpretation, being one key reason for the comparably “superficial” results of many automated approaches to date compared to manual content analyses (see Sect. 2.5 ). For example, determining implicitly ascribed frame properties, such as “reason,” requires a high degree of interpretation since typically a news articles would not state that a person acted reasonably but instead this conclusion would be made by news consumers after reading one or more sentences describing, for example, actions that portray the person in a specific way. One idea to improve the classification performance is to use deep language models, such as RoBERTa, pre-trained on large amounts of also news articles [ 214 ]. RoBERTa and other language models have significantly improved natural language understanding capabilities across many tasks [ 373 ]. Given these advancements, we expect that such language models can also reliably determine complex and implicitly ascribed frame properties. However, fine-tuning language models, especially for multi-label classification tasks with high degree of required interpretation as for frames and frame properties, require also the creation of very large datasets (cf. [ 46 ]).

The second cause is the difficulty of frame annotation as well as frame classification or in our cases more specifically the effects of framing, i.e., frame properties. As our comparably low inter-coder reliability (see Sect. 5.2.2 ) and prior work indicate [ 45 ], the annotation of frames and frame properties is highly complex, and some “degree of subjectivity in framing analysis [is] unavoidable” [ 45 ]. In our view, the most effective idea to address the high annotation difficulty is to reduce the number of frame properties that are to be annotated. Other, commonly used means to tackle the subjectivity are performing more training iterations, might be less effective since we already conducted as many iterations as we could to improve the inter-coder reliability. Another promising idea is to determine frame properties on a much larger set of news articles. While our exploratory evaluation showed in principle also expected framing patterns, we tested our method only on five articles for each use case. We think that the task’s ambiguity could—besides technical improvements as mentioned previously—be addressed by identifying framing patterns on more articles instead of attempting to pinpoint frame properties on individual articles.

5.2.5 Conclusion

This section presented the results of our exploratory research on imitating manual frame analysis. Albeit effective, such analysis entails the definition of topic-specific and analysis question-specific frames. Such dependencies are in contrast to the objectives of the person-oriented framing analysis approach, which is intended to be applied to any news coverage on policy issues. Instead of frames, we proposed to analyze frame properties, which represent the effects of person-oriented framing, such as whether a person is shown as being “aggressive.”

In our view, the approach represents a promising line of research but at the same time suffers from shortcomings that are common to prior approaches aiming to imitate frame analysis, especially high annotation cost for the required training dataset. Likewise, we noticed a degree of subjectivity that could not be reduced without lowering the “substance” of the frame properties (cf. [ 45 ]), which is required to interpret the “meaning between the lines” like it is done in frame analysis.

5.3 Target-Dependent Sentiment Classification

This section describes the second approach for our frame analysis component. Specifically, we describe a dataset and method for target-dependent sentiment classification (TSC, also called aspect-term sentiment classification) in news articles. In the context of the overall person-oriented framing analysis (PFA) and in particular the frame analysis component, we use TSC to classify a fundamental effect of person-oriented framing, i.e., whether sentences and articles portray individual persons positively or negatively. As we show in this section and our prototype evaluation (Chap. 6 ), TSC represents a pragmatic and effective alternative to the fine-grained but expensive approach of classifying frame properties. The advantages of TSC over approaches aiming to capture frames or frame derivatives are the reduced annotation cost and high reliability.

We define our objective in this section as follows: we seek to detect polar judgments toward target persons [ 335 ]. Following the TSC literature, we include only in-text, specifically in-sentence, means to express sentiment. In news texts, such means are, for example, word choice or describing actions performed by the target, e.g., “John and Bert got in a fight” or “John attacked Bert.” Sentiment can also be expressed indirectly, e.g., through quoting another person, such as “According to John, an expert on the field, the idea ‘suffers from fundamental issues’ such as […]” [ 335 ]. Other means may also alter the perception of persons and topics in the news, but are not in the scope of the task [ 16 ], e.g., because they are not on sentence level, for example, story selection, source selection, article’s placement and size (Sect. 2.1 ), and epistemological bias [ 297 ]. Albeit excluding non-sentence-level means from our objective in this section, in the context of the overall thesis, the TSC method will still be able to catch the effects of source selection and commission and omission of information. For example, when journalists write articles and include mostly information of one perspective that is in favor of specific persons, the resulting article will mostly reflect that perspective and thus be in favor of these persons (Sect. 3.3.2 ).

The main contributions of this section are as follows: (1) We create a small-scale dataset and train state-of-the-art models on it to explore characteristics of sentiment in news articles. (2) We introduce NewsMTSC , a large, manually annotated dataset for TSC in political news articles. We analyze the quality and characteristics of the dataset using an on-site, expert annotation. Because of its fundamentally different characteristics compared to previous TSC datasets, e.g., as to how sentiment is expressed and text lengths, NewsMTSC represents a challenging novel dataset for the TSC task. (3) We propose a neural model that improves TSC performance on news articles compared to prior state-of-the-art models. Additionally, our model yields competitive performance on established TSC datasets. (4) We perform an extensive evaluation and ablation study of the proposed model. Among others, we investigate the recently claimed “degeneration” [ 161 ] of TSC to sequence-level classification, finding a performance drop in all models when comparing single- and multi-target sentences.

The remainder of this section is structured as follows. In Sect. 5.3.1 , we provide an overview of related work and identify the research gap of sentiment classification in news articles. In Sect. 5.3.2 , we explore the characteristics of how sentiment is expressed in news articles by creating and analyzing a small-scale TSC dataset. We then use and address the findings of this exploratory work, to create our main dataset (Sect. 5.3.3 ) and model (Sect. 5.3.4 ). Key differences and improvements of the main dataset compared to the small-scale dataset are as follows. We significantly increase the dataset’s size and the number of annotators per example and address class imbalance. Further, we devise annotation instructions specifically created to capture a broad spectrum of sentiment expressions specific to news articles. In contrast, the early dataset misses the more implicit sentiment expressions commonly used by news authors (see Sect. 5.3.2.5 ). Also, we test various consolidation strategies and conduct an expert annotation to validate the dataset.

We provide the dataset and code to reproduce our experiments at

https://github.com/fhamborg/NewsMTSC .

5.3.1 Related Work

Analogously to other NLP tasks, the TSC task has recently seen a significant performance leap due to the rise of language models [ 73 ]. Pre-BERT approaches yield up to F 1 m  = 63.3 on the SemEval 2014 Twitter set [ 182 ]. They employ traditional machine learning combining hand-crafted sentiment dictionaries, such as SentiWordNet [ 13 ], and other linguistic features [ 29 ]. On the same dataset, vanilla BERT (also called BERT-SPC) yields 73.6 [ 73 , 400 ]. Specialized downstream architectures improve performance further, e.g., LCF-BERT yields 75.8 [ 400 ].

The vast majority of recently proposed TSC approaches employ BERT and focus on devising specialized downstream architectures [ 329 , 346 , 400 ]. More recently, to improve performance further, additional measures have been proposed, for example, domain adaption of BERT, i.e., domain-specific language model fine-tuning prior to the TSC fine-tuning [ 76 , 300 ]; use of external knowledge, such as sentiment or emotion dictionaries [ 151 , 401 ], rule-based sentiment systems [ 151 ], and knowledge graphs [ 102 ]; use of all mentions of a target and/or related targets in a document [ 50 ]; and explicit encoding of syntactic information [ 286 , 398 ].

To train and evaluate recent TSC approaches, three datasets are commonly used: Twitter [ 257 , 258 , 305 ], Laptop, and Restaurant [ 289 , 290 ]. These and other TSC datasets [ 273 ] suffer from at least one of the following shortcomings. First, implicitly or indirectly expressed sentiment is rare in them. In their domains, e.g., social media and reviews, typically, authors explicitly express their sentiment regarding a target [ 402 ]. Second, they largely neglect that a text may contain coreferential mentions of the target or mentions of different concepts (with potentially different polarities), respectively [ 161 ].

Texts in news articles differ from reviews and social media in that news authors typically do not express sentiment toward a target explicitly (exceptions include opinion pieces and columns). Instead, journalists implicitly or indirectly express sentiment (Sect. 2.3.4 ) because language in news is typically expected to be neutral and journalists to be objective [ 16 , 110 ].

Our objective as described in the beginning of Sect. 5.3 is largely identical to prior news TSC literature [ 16 , 335 ] with key differences: we do not generally discard the “author level” and “reader level.” Doing so would neglect large parts of sentiment expressions. Thus, it would degrade real-world performance of the resulting dataset and models trained on it. For example, word choice (listed as “author level” and discarded from their problem statement) is in our view an in-text means that may in fact strongly influence how readers perceive a target, e.g., “compromise” or “consensus.” While we do not exclude the “reader level,” we do seek to exclude polarizing or contentious cases, where no uniform answer can be found in a set of randomly selected readers (Sects. 5.3.3.3 and 5.3.3.4 ). As a consequence, we generally do not distinguish between the three levels of sentiment (“author,” “reader,” and “text”).

Previous news TSC approaches mostly employ sentiment dictionaries, e.g., created manually [ 16 , 335 ] or extended semi-automatically [ 110 ], but yield poor or even “useless” [ 335 ] performances. To our knowledge, there exist two datasets for the evaluation of news TSC methods. Steinberger et al. [ 335 ] proposed a news TSC dataset, which—perhaps due to its small size ( N  = 1274)—has not been used or tested in recent TSC literature. Another dataset contains quotes extracted from news articles, since quotes more likely contain explicit sentiment ( N  = 1592) [ 16 ].

In summary, no suitable datasets for news TSC exist nor have news TSC approaches been proposed that exploit recent advances in NLP.

5.3.2 Exploring Sentiment in News Articles

We describe how the procedure used to create our exploratory TSC dataset for the domain of news articles, including the collection of articles and the annotation procedure. Afterward, we discuss the characteristics of the dataset. Then, we report the results of our evaluation where we test state-of-the-art TSC models on the dataset. Lastly, we discuss the findings and shortcomings of our qualitative dataset investigation and quantitative evaluation to derive means to address these shortcomings in our main dataset.

5.3.2.1 Creating an Exploratory Dataset

Our procedure to create the exploratory dataset for sentiment classification on news articles entails the following steps. First, we create a base set of articles of high diversity in topics covered and writing styles, e.g., whether emotional or factual words are used (cf. [ 96 ]). Using our news extractor (Sect. 3.5 ), we collect news articles from the Common Crawl news crawl (CCNC, also known as CC-NEWS), consisting of over 250M articles until August 2019 [ 256 ]. To ensure diversity in writing styles, we select 14 US news outlets, Footnote 2 which are mostly major outlets that represent the political spectrum from left to right, based on selections by Budak et al. [ 38 ], Groseclose and Milyo [ 117 ], and Baum and Groeling[ 21 ]. We cannot simply select the whole corpus, because CCNC lacks articles for some outlets and time frames. By selecting articles published between August 2017 and July 2019, we minimize such gaps while covering a time frame of 2 years, which is sufficiently large to include many diverse news topics. To facilitate the balanced contribution of each outlet and time range, we perform binning: we create 336 bins, one for each outlet and month, and randomly draw 10 articles reporting on politics for each bin, resulting in 3360 articles in total. Footnote 3 During binning, we remove any article duplicates by text equivalence.

To create examples for annotation, we select all mentions of NEs recognized as PERSON, NROP, or ORG for each article [ 376 ]. Footnote 4 We discard NE mentions in sentences shorter than 50 characters. For each NE mention, we create an example by using the mention as the target and its surrounding sentence as its context. We remove any example duplicates. Afterward, to ensure diversity in writing styles and topics, we use the outlet-month binning described previously and randomly draw examples from each bin.

Different means may be used to address expected class imbalance, e.g., for the Twitter set, only examples that contained at least one word from a sentiment dictionary were annotated [ 257 , 258 ]. While doing so yields high frequencies of classes that are infrequent in real-world distribution, it also causes dataset shift and selection bias [ 293 ]. Thus, we instead investigate the effectiveness of different means to address class imbalance during training and evaluation (see Sect. 5.3.2.4 ).

5.3.2.2 Annotating the Exploratory Dataset

We set up an annotation process following best practices from TSC literature [ 258 , 290 , 305 , 335 ]. For each example, we asked three coders to read the context, in which we visually highlighted the target, and assess the target’s sentiment. Examples were shown in random order to each coder. Coders could choose from positive , neutral , and negative polarity, whereby they were allowed to choose positive and negative polarity at the same time. Coders were asked to reject an example, e.g., if it was not political or a meaningless text fragment. Before, coders read a codebook that included instructions on how to code and examples. Five coders, students, aged between 24 and 32, participated in the process.

In total, 3288 examples were annotated, from which we discard 125 (3.8%) that were rejected by at least one coder, resulting in 3163 non-rejected examples. From these, we discard 3.3% that lacked a majority class, i.e., examples where each coder assigned a different sentiment class, and 1.8% that were annotated as positive and negative sentiment at the same time, to allow for better comparison with previous TSC datasets and methods (see Sect. 5.3.1 ). Lastly, we split the remaining 3002 examples into training and test sets; see Table 5.4 .

We use the full set of 3163 non-rejected examples to illustrate the degree of agreement between coders: 3.3% lack a majority class; for 62.7%, two coders assigned the same sentiment; and for 33.9%, all coders agreed. On average, the accuracy of individual coders is A h  = 72.9%. We calculate two inter-rater reliability (IRR) measures. For completeness, Cohen’s kappa is κ  = 25.1, but it is unreliable in our case due to Kappa’s sensitivity to class imbalance [ 55 ]. The mean pairwise observed agreement over all coders is 72.5.

5.3.2.3 Exploring the Characteristics of Sentiment in News Articles

In a manual, qualitative analysis of our exploratory dataset, we found two key differences of news compared to established domains: First, we confirmed that news contains mostly implicit and indirect sentiment (see Sect. 5.3.1 ). Second, determining the sentiment in news articles typically requires a greater degree of interpretation (cf. [ 335 ]). The second difference is caused by multiple factors, particularly the implicitness of sentiment (mentioned as the first difference) and that sentiment in news articles is more often dependent on non-local, i.e., off-sentence, context. In the following, we discuss annotated examples (part of the dataset and discarded examples) to understand the characteristics of target-dependent sentiment in news texts.

In our analysis, we found that in news articles, a key means to express targeted sentiment is to describe actions performed by the target. This is in contrast, e.g., to product reviews where more often a target’s feature, e.g., “high resolution,” or the mention of the target itself, e.g., “the camera is awesome,” expresses sentiment. For example, in “The Trump administration has worked tirelessly to impede a transition to a green economy with actions ranging from opening the long-protected Arctic National Wildlife Refuge to drilling […],” the target (underlined) was assigned negative sentiment due to its actions.

We found sentiment in ≈3% of the examples to be strongly reader-dependent (cf. [ 16 ]). Footnote 5 In the previous example, the perceived sentiment may, in part, depend on the reader’s own ideological or political stance, e.g., readers focusing on economic growth could perceive the described action positively, whereas those concerned with environmental issues would perceive it negatively.

In some examples, targeted sentiment expressions can be interpreted differently due to ambiguity. As a consequence, we mostly found such examples in the discarded examples, and thus they are not contained in our exploratory dataset. While this can be true for any domain (cf. “polarity ambiguity” in [ 290 ]), we think it is especially characteristic for news articles, which are lengthier than tweets and reviews, giving authors more ways to refer to non-local statements and to embed their arguments in larger argumentative structures. For instance, in “And it is true that even when using similar tactics, President Trump and President Obama have expressed very different attitudes towards immigration and espoused different goals,” the target was assigned neutral sentiment. However, when considering this sentence in the context of its article [ 356 ], the target’s sentiment may be shifted (slightly) negatively.

From a practical perspective, considering more context than only the current sentence seems to be an effective means to determine otherwise ambiguous sentiment expressions. By considering a broader context, e.g., the current sentence and previous sentences, annotators can get a more comprehensive understanding of the author’s intention and the sentiment the author may have wanted to communicate. The greater degree of interpretation required to determine non-explicit sentiment expressions may naturally lead to a higher degree of subjectivity. Due to our majority-based consolidation method (see Sect. 5.3.2.2 ), examples with non-explicit or apparently ambiguous sentiment expressions are not contained in our exploratory dataset.

5.3.2.4 Experiments and Discussion

We evaluated three TSC methods that represent the state of the art on the established TSC datasets Laptop, Restaurant, and Twitter: AEN-BERT [ 329 ], BERT-SPC [ 73 ], and LCF-BERT [ 400 ]. Additionally, we tested the methods using a domain-adapted language model, which we created by fine-tuning BERT (base, uncased) for 3 epochs on 10M English sentences sampled from CCNC (cf. [ 300 ]). For all methods, we test hyperparameter ranges suggested by their respective authors. Footnote 6 Additionally, we investigated the effects of two common measures to address class imbalance: weighted cross-entropy loss (using inverse class frequencies as weights) and oversampling of the training set. Of the training set, we use 2001 examples for training and 300 for validation.

We used average recall ( R a ) as our primary measure, which was also chosen as the primary measure in the TSC task of the latest SemEval series, due to its robustness against class imbalance [ 305 ]. We also measured accuracy ( A ), macro F1 ( F 1 m ), and average F1 on positive and negative classes ( F 1 pn ) to allow comparison to previous works [ 258 ].

Table 5.5 shows that LCF-BERT performed best ( R a  = 67.3 using BERT and 69.8 using our news-adapted language model). Footnote 7 Class-weighted cross-entropy loss helped best to address class imbalance ( R a  = 69.8 compared to 67.2 using oversampling and 64.6 without any measure).

Performance in news articles was significantly lower than in established domains, where the top model (LCF-BERT) yielded in our experiments R a  = 78.0 (Laptop), 82.2 (Restaurant), and 75.6 (Twitter). For Laptop and Restaurant, we used domain-adapted language models [ 300 ]. News TSC accuracy A  = 66.0 was lower than single human level A h  = 72.9 (see Sect. 5.3.2.3 ).

We carried out a manual error analysis (up to 30 randomly sampled examples for each true class). We found target misassociation as the most common error cause: In 40%, sentences express the predicted sentiment toward a different target. In 30%, we cannot find any apparent cause. The remaining cases contain various potential causes, including usage of euphemisms or sayings (12% of examples with negative sentiment). Infrequently, we found that sentiment is expressed by rare words or figurative speech or is reader-dependent (the latter in 2%, approximately matching the 3% of reader-dependent examples reported in Sect. 5.3.2.3 ).

Previous news TSC approaches, mostly dictionary-based, could not reliably classify implicit or indirect sentiment expressions (see Sect. 5.3.1 ). In contrast, our experiments indicate that BERT’s language understanding suffices to interpret implicitly expressed sentiment correctly (cf. [ 16 , 73 , 110 ]). Our exploratory dataset does not contain instances in which the broader context defines sentiment, since human coders could or did not classify them in our annotation procedure. Our experiments therefore cannot elucidate this particular characteristic discussed in Sect. 5.3.2.3 .

5.3.2.5 Summary

We explored how target-dependent sentiment classification (TSC) can be applied to political news articles. After creating an exploratory dataset of 3000 manually annotated sentences sampled from news articles reporting on policy issues, we qualitatively analyzed its characteristics. We found notable differences concerning how authors express sentiment toward targets as compared to other, well-researched domains of TSC, such as product reviews or posts on social media. In these domains, authors tend to explicitly express their opinions. In contrast, in news articles, we found dominant use of implicit or indirect sentiment expressions, e.g., by describing actions, which were performed by a given target, and their consequences. Thus, sentiment expressions may be more ambiguous, and determining their polarity requires a greater degree of interpretation.

In our quantitative evaluation, we found that current TSC methods performed lower on the news domain (average recall R a  = 69.8 using our news-adapted BERT model, R a  = 67.3 without) than on popular TSC domains ( \(R_a=\left [ 75.6, 82.2\right ]\) ).

While our exploratory dataset contains clear sentiment expressions, it lacks other sentiment types that occur in real-world news coverage, for example, sentences that express sentiment more implicitly or ambiguously. To create a labeled TSC dataset that better reflects real-world news coverage, we suggest to adjust annotation instructions to raise annotators’ awareness of these sentiment types and clearly define how they should be labeled. Technically, apparently ambiguous sentiment expressions might be easier to label when considering a broader context, e.g., not only the current sentence but also previous sentences. Considering more context might also help to improve a classifier’s performance.

5.3.3 NewsMTSC: Dataset Creation

This section describes the procedure to create our main dataset for TSC in the news domain. When creating the dataset, we rely on best practices reported in literature on the creation of datasets for NLP [ 291 ], especially for the TSC task [ 305 ]. As our previous exploration has showed (Sect. 5.3.2.5 ), compared to previous TSC datasets though, the nature of sentiment in news articles requires key changes, especially in the annotation instructions and consolidation of answers [ 335 ].

5.3.3.1 Data Sources

We use two datasets as sources: our POLUSA dataset [ 96 ] and the Bias Flipper 2018 (BF18) dataset [ 49 ]. Both satisfy five criteria that are important to our problem. First, they contain news articles reporting on political topics. Second, they approximately match the online media landscape as perceived by an average US news consumer. Footnote 8 Third, they have a high diversity in topics due to the number of articles contained and time frames covered (POLUSA: 0.9M articles published between Jan. 2017 and Aug. 2019, BF18: 6447 articles associated with 2781 events). Fourth, they feature high diversity in writing styles because they contain articles from across the political spectrum, including left- and right-wing outlets. Fifth, we find that they contain only few minor content errors albeit being created through scraping or crawling.

In early tests when selecting data sources, we tested other datasets as well. While we found that other factors are more important for the resulting quality of annotated examples (filtering of candidate example, annotation instructions, and consolidation strategy), we also found that other datasets are slightly less suitable as to the five previously mentioned criteria because the datasets, e.g., contain only contentious news topics and articles [ 45 ] or hyperpartisan sentences [ 178 ], are of mixed content quality [ 264 ] or contain too few sentences [ 4 , 5 ].

5.3.3.2 Creation of Examples

To create a batch of examples for annotation, we devise a three tasks process: First, we extract example candidates from randomly selected articles. Second, we discard non-optimal candidates. Only for the train set, third, we filter candidates to address class imbalance. We repeatedly execute these tasks so that each batch yields 500 examples for annotation, contributed equally by both sources.

First, we randomly select articles from the two sources. Since both are at least very approximately uniformly distributed over time [ 49 , 96 ], randomly drawing articles will yield sufficiently high diversity in both writings styles and reported topics (Sect. 5.3.3.1 ). To extract from an article examples that contain meaningful target mentions, we employ coreference resolution (CR). Footnote 9 We iterate all resulting coreference clusters of the given article and create a single example for each mention and its enclosing sentence.

Extraction of mentions of named entities (NEs) is the commonly employed method to create examples in previous TSC datasets [ 257 , 258 , 305 , 335 ]. We do not use it since we find it would miss \(\gtrapprox \) 30% mentions of relevant target candidates, e.g., pronominal or near-identity mentions.

Second, we perform a two-level filtering to improve quality and “substance” of candidates. On coreference cluster level, we discard a cluster c in a document d if | M c |≤ 0.2| S d |, where |…| is the number of mentions of a cluster ( M c ) and sentences in a document ( S d ). Also, we discard non-persons clusters, i.e., if ∃ m  ∈  M c  :  t ( m )∉{−, P }, where t ( m ) yields the NE type Footnote 10 of m and − and P represent the unknown and person type, respectively. On example level, we discard short and similar examples e , i.e., if \(|s_{e}| < 50 \lor \exists \hat {e}: \mathrm {sim}(s_{e},s_{\hat {e}} )>0.6 \land m_e=m_{\hat {e}} \land t_{e} = t_{\hat {e}}\) where s e , m e , and t e are the sentence of e , its mention, and the target’s cluster, respectively, and sim(…) is the cosine similarity. Lastly, if a cluster has multiple mentions in a sentence, we try to select the most meaningful example. In short, we prefer the cluster’s representative mention Footnote 11 over nominal mentions and those over all other instances.

Third, for only the train set, we filter candidates to address class imbalance. Specifically, we discard examples e that are likely the majority class ( p (neutral| s e ) > 0.95) as determined by a simple binary classifier [ 310 ]. Whenever annotated and consolidated examples are added to the train set of NewsMTSC, we retrain the classifier on them and all previous examples in the train set.

5.3.3.3 Annotation

Instructions used in popular TSC datasets plainly ask annotators to rate the sentiment of a text toward a target [ 290 , 305 ]. For news texts, we find that doing so yields two issues [ 16 ]: low inter-rater reliability (IRR) and low suitability. Low suitability refers to examples where annotators’ answers can be consolidated but the resulting majority answer is incorrect as to the task. For example, instructions from prior TSC datasets often yield low suitability for polarizing targets, independent of the sentence they are mentioned in. Figure 5.1 depicts our final annotation instructions.

figure 1

Final version of the annotation instructions as shown on Amazon Mechanical Turk

In an interactive process with multiple test annotations (six on-site and eight on Amazon Mechanical Turk, MTurk), we test various measures to address the two issues. We find that asking annotators to think from the perspective of the sentence’s author strongly facilitates that annotators overcome their personal attitude. Further, we find that we can effectively draw annotators’ attention not only at the event and other “facts” described in the sentence (the “what”) but also at word choice (“how” it is described) by exemplarily mentioning both factors and abstracting these factors as the author’s holistic “attitude.” Footnote 12 We further improve IRR and suitability, e.g., by explicitly instructing annotators to rate sentiment only regarding the target but not regarding other aspects, such as the reported event.

We also test other means to address low IRR and suitability in news TSC annotation but find our means to be more efficient while similarly effective. For example, Balahur et al. [ 16 ] ask annotators to only rate the target’s sentiment but not consider whether the news are “good” or “bad.” They also ask annotators to interpret only “what is said” and not use their own background knowledge. Additionally, we test a design where we ask annotators to select the more negative sentence of a pair of sentences sharing a target. We use semantic textual similarity (STS) datasets [ 4 , 5 ] and extract all pairs with an STS score >2.5. While this design yields high IRR, suitability (especially political framing through word choice is found more effectively [ 166 ]), and efficiency, the STS datasets contain too few examples. On MTurk, we find consistently across all instruction variants that short instructions yield higher suitability and IRR than more comprehensive instructions. Surprisingly, the average duration of each crowdworkers’ first assignment is shorter for the latter. This is perhaps because crowdworkers have high incentive to minimize the duration per task to increase their salary and in case they deem instructions too long, the crowdworkers will not read them at all or only very briefly [ 302 , 322 ].

While most TSC dataset creation procedures use 3- or 5-point Likert scales [ 16 , 257 , 258 , 289 , 290 , 305 , 335 ], we use a 7-point scale to encourage rating also only slightly positive or negative examples as such.

Technically, we closely follow previous literature on TSC datasets [ 290 , 305 ]. We conduct the annotation of our examples on MTurk. Each example is shown to five randomly selected crowdworkers. To participate in our annotation, crowdworkers must have the “Master” qualification, i.e., have a record of successfully completed, high-quality work on MTurk. To ensure quality, we implement a set of objective measures and tests [ 180 ]. While we pay all crowdworkers always (USD 0.07 per assignment), we discard all of a crowdworker’s answers if at least one of the following conditions is met. (a) A crowdworker was not shown any test question or answered at least one incorrectly, Footnote 13 (b) a crowdworker provided answers to invisible fields in the HTML form (0.3% of crowdworkers did so, supposedly bots), or (c) the average duration of time spent on the assignments was extremely low (<4 s ).

The IRR is sufficiently high ( κ C  = 0.74) when considering only examples in NewsMTSC. The expected mixed quality of crowdsourced work becomes apparent when considering all examples, including those that could not be consolidated and answers of those crowdworkers who did not pass our quality checks ( κ C  = 0.50).

5.3.3.4 Consolidation

We consolidate the answers of each example to a majority answer by employing a restrictive strategy. Specifically, we consolidate the set of five answers A to the single-label three-class polarity p  ∈{pos., neu., neg.} if ∃ C  ⊆  A  : | C |≥ 4 ∧∀ c  ∈  C  :  s ( c ) =  p , where s ( c ) yields the three-class polarity of an individual seven-class answer c , i.e., neutral ⇒ neutral, any positive (from slightly to strongly) ⇒ positive, and, respectively, for negative. If there is no such consolidation set C , A cannot be consolidated, and the example is discarded. Consolidating to three-class polarity allows for direct comparison to established TSC dataset.

While the strategy is restrictive (only 50.6% of all examples are consolidated this way), we find it yields the highest quality. We quantify the dataset’s quality by comparing the dataset to an expert annotation (Sect. 5.3.3.6 ) and by training and testing models on dataset variants with different consolidations. Compared to consolidations employed for previous TSC datasets, quality is improved significantly on our examples, e.g., our strategy yields F 1 m  = 86.4 when compared to experts’ annotations and models trained on the resulting set yield up to F 1 m  = 83.1, whereas the two-step majority strategy employed for the Twitter 2016 set [ 258 ] yields 50.6 and 53.4, respectively.

5.3.3.5 Splits and Multi-Target Examples

NewsMTSC consists of three sets as depicted in Tables 5.6 and 5.7 . For the train set, we employ class balancing prior to annotation (Sect. 5.3.3.2 ). To minimize dataset shift, which might yield a skewed sentiment distribution in the dataset compared to the real world [ 293 ], we do not use class balancing for either of the two test sets. Sentences can have multiple targets (MT) with potentially different polarities. We call this MT property . To investigate the effect on TSC performance of considering or neglecting the MT property [ 161 ], we devise a test set named test-mt , which consists only of examples that have at least two semantically different targets, i.e., each belonging to a separate coreference cluster (Sect. 5.3.3.2 ). Since the additional filtering required for test-mt naturally yields dataset shift, we create a second test set named test-rw , which omits the MT filtering and is thus designed to be as close as possible to the real-world distribution of sentiment. We seek to provide a sentiment score for each person in each sentence in train and test-rw , but mentions may be missing, e.g., because of erroneous coreference resolution or crowdworkers’ answers could not be consolidated. Table 5.7 shows the frequencies of the targets and sentiment classes with added coreferential mentions.

5.3.3.6 Quality and Characteristics

We conducted an expert annotation of a random subset of 360 examples used during the creation of NewsMTSC with 5 international graduate students (studying Political or Communication Science at the University of Zurich, Switzerland, 3 female, 2 male, aged between 23 and 29). Key differences compared to the MTurk annotation are as follows: First is extensive training until high IRR is reached (considering all examples, κ C  = 0.72; only consolidated, κ C  = 0.93). We conducted five iterations, each consisting of individual annotations by the students, quantitative and qualitative review, adaption of instructions, and individual and group discussions. Second are comprehensive instructions (4 pages). Third is no time pressure, since the students were paid per hour (crowdworkers per assignment).

When comparing the expert annotation with our dataset, we found that NewsMTSC is of high quality ( F 1 m  = 86.4). The quality of unfiltered answers from MTurk is, as expected, much lower (50.1).

What is contained in NewsMTSC? In a random set of 50 consolidated examples from MTurk, we found that most frequent, non-mutually exclusive means to express a polar statement (62% of the 50) are usage of quotes (in total, direct, and indirect 42%, 28%, and 14%, respectively), target being subject to action (24%), evaluative expression by the author or an opinion holder mentioned outside of the sentence (18%), target performing an action (16%), and loaded language or connotated terms (14%). Direct quotes often contain evaluative expressions or connotated terms and indirect quotes less. Neutral examples (38% of the 50) contain mostly objective storytelling about neutral events (16%) or variants of “[target] said that […]” (8%). Yet, “said” variants cannot be used as a reliably indicator for neutral sentiment, e.g., if the target has multiple mentions in the sentences or if the target’s statement is considered positive or negative, e.g., “‘Not all of that is preventable, but a lot of it is preventable if we’ve got better cooperation […],’ Obama said.”

What is not contained in NewsMTSC? We qualitatively reviewed all examples where individual answers could not be consolidated to identify potential causes why annotators do not agree. The predominant reason is technical, i.e., the restrictiveness of the consolidation (MTurk compared to experts: 26% ≈ 30%). Other examples lack apparent causes (24% ≫ 8%). Further potential causes are (not mutually exclusive) as follows: ambiguous sentence (16% ≈ 18%), sentence contains positive and negative parts (8% ≈ 6%), and opinion holder is target (6% ≈ 8%), e.g., “[…] Bauman asked supporters to ‘push back’ against what he called a targeted campaign to spread false rumors about him online.” In a subset of such instances, more context could have helped to resolve ambiguity, e.g., by showing annotators also the sentence prior to the mention.

What are qualitative differences in the annotations by crowdworkers and experts? We reviewed all 63 cases (18%) where answers from MTurk could be consolidated but differ to experts’ answers. The major reason for disagreement is the restrictiveness of the consolidation (53 cases have no consolidation among the experts). In ten cases, the consolidated answers differ. We found that in few examples (2–3%), crowdsourced annotations are superficial and fail to interpret the full sentence correctly.

Texts in NewsMTSC are much longer than in prior TSC datasets (mean over all examples): 152 characters compared to 100, 96, and 90 in Twitter, Restaurant, and Laptops, respectively.

5.3.4 Method

The goal of TSC is to find a target’s polarity y  ∈{pos., neu., neg.} in a sentence. Our model consists of four key components (Fig. 5.2 ): a pre-trained language model (LM), a representation of external knowledge sources (EKS), a target mention mask, and a bidirectional GRU (BiGRU) [ 51 ]. We adapt our model from Hosseinia, Dragut, and Mukherjee [ 151 ] and change the design as follows: we employ a target mask (which they did not) and use multiple EKS simultaneously (instead of one). Further, we use a different set of EKS (Sect. 5.3.5 ) and do not exclude the LM’s parameters from fine-tuning.

figure 2

Architecture of the proposed model for target-dependent sentiment classification

Input Representation

We construct three model inputs. The first is a text input T constructed as suggested by Devlin et al. [ 73 ] for question answering (QA) tasks. Specifically, we concatenate the sentence and target mention and tokenize the two segments using the LM’s tokenizer and vocabulary, e.g., WordPiece for BERT [ 386 ]. Footnote 14 This step results in a text input sequence \(T=[\mathrm {CLS}, s_{0}, s_{1}, \ldots , s_{p}, \mathrm {SEP}, t_0, t_1, \ldots , t_q, \mathrm {SEP}] \in \mathbb {N}^{n}\) consisting of n word pieces, where n is the manually defined maximum sequence length.

Various forms of this representation have been proposed, e.g., opposite order sentence and target or instead of the plain target mention using a natural language question or pseudo sentence [ 151 , 346 ]. We find that on average in the TSC domain, they yield lower performance than the plain variant that we employ.

The second input is a feature representation of the sentence, which we create using one or more EKS, such as dictionaries [ 151 , 401 ]. Given an EKS with d dimensions, we construct an EKS representation \(E \in \mathbb {R}^{n\times d}\) of S , where each vector e i ∈{0,1,…, p } is a feature representation of the word piece i in the sentence. For example, when using a sentiment dictionary with two mutually non-exclusive polarities’ dimensions positive and negative [ 153 ], d  = 2. Given a sentence “Good […],” we set e 1  = [1, 0]. To facilitate learning associations between the token-based EKS representation and the WordPiece-based sequence T , we create E so that it contains k repeated vectors for each token where k is the token’s number of word pieces. Thereby, we also consider special characters, such as CLS. If multiple EKS with a total number of dimensions \(\hat {d} = \sum d\) are used, their representations of the sentence are stacked resulting in \(E \in \mathbb {R}^{n\times \hat {d}}\) .

The third input is a target mask \(M \in \mathbb {R}^{n}\) , i.e., for each word piece i in the sentence that belongs to the target, m i  = 1, else 0 [ 94 ].

Embedding Layer

We feed T into the LM to yield a contextualized word embedding of shape \(\mathbb {R}^{n\times h}\) , where h is the number of hidden states in the language model, e.g., h  = 768 for BERT [ 73 ]. We feed E into a randomly initialized matrix \(W_E \in \mathbb {R}^{\hat {d} \times h}\) to yield an EKS embedding. We repeat M to be of shape \(\mathbb {R}^{n \times h}\) . By creating all embeddings in the same shape, we facilitate a balanced influence of each input to the model’s downstream components. We stack all embeddings to form a matrix \([TEM] \in \mathbb {R}^{n\times 3h}\) .

Interaction Layer

We allow the three embeddings to interact using a single-layer BiGRU [ 151 ], which yields hidden states \(H \in \mathbb {R}^{n\times 6h} = \mathrm {BiGRU}([TEM])\) . RNNs, such as LSTMs and GRUs, are commonly used to learn a higher-level representation of a word embedding, especially in state-of-the-art TSC prior to BERT-based models but also recently [ 151 , 208 , 213 , 401 ]. We choose an BiGRU over an LSTM because of the smaller number of parameters in BiGRUs, which may in some cases result in better performance [ 54 , 118 , 151 , 161 ].

Pooling and Decoding

We employ three common pooling techniques to turn the interacted, sequenced representation H into a single vector [ 151 ]. We calculate element-wise (1) mean and (2) maximum over all hidden states H and retrieve the (3) last hidden state h n −1 . Then, we stack the three vectors to P , feed P into a fully connected layer FC so that z  =  FC ( P ), and calculate y  =  σ ( z ).

5.3.5 Evaluation

This section describes the experiments we conducted to evaluate our model for target-dependent sentiment classification.

Data and Metrics

In addition to NewsMTSC, we used the three established TSC sets: Twitter, Laptop, and Restaurant. We used metrics established in the TSC literature: macro F1 on all ( F 1 m ) and only the positive and negative classes ( F 1 pn ), accuracy ( A ), and average recall ( R a ). If not otherwise noted, performances are reported for our primary metric, F 1 m .

We compared our model with TSC methods that yield state-of-the-art results on at least one of the established datasets: SPC-BERT [ 73 ]: input is identical to our text input. FC and softmax are calculated on CLS token. TD-BERT [ 94 ]: masks hidden states depending on whether they belong to the target mention. LCF-BERT [ 400 ]: similar to TD but additionally weights hidden states depending on their token-based distance to the target mention. We used the improved implementation [ 394 ] and enable the dual-LM option, which yields slightly better performance than using only one LM instance [ 400 ]. We also planned to test LCFS-BERT [ 286 ], but due to technical issues, we were not able to reproduce the authors’ results and thus exclude LCFS from our experiments.

Implementation Details

To find for each model the best parameter configuration, we performed an exhaustive grid search. Any number we report is the mean of five experiments that we run per configuration. We randomly split each test set into a dev-set (30%) and the actual test-set (70%). We tested the base version of three LMs: BERT, RoBERTa, and XLNET. For all methods, we tested parameters suggested by their respective authors. Footnote 15 We tested all 15 combinations of the following 4 EKS: (1) SENT [ 153 ], a sentiment dictionary (number of non-mutually exclusive dimensions, 2; domain, customer reviews); (2) LIWC [ 357 ], a psychometric dictionary (73, multiple); (3) MPQA [ 383 ], a subjectivity dictionary (3, multiple); and (4) NRC [ 247 ], dictionary of sentiment and emotion (10, multiple).

Overall Performance

Table 5.8 reports the performances of the models using different LMs and evaluated on both test sets. We found that the best performance was achieved by our model ( F 1 m  = 83.1 on test-rw compared to 81.8 by the prior state of the art). For all models, performances were improved when using RoBERTa, which is pre-trained on news texts, or XLNET, likely because of its large pre-training corpus. XLNET is not reported in Table 5.8 since its performances were generally similar to those of RoBERTa except for the TD model, where XLNET degrades performance by 5–9pp. Looking at BERT, we found no significant improvement of the proposed model over the prior state of the art. Even if we domain-adapted BERT [ 300 ] for 3 epochs on a random sample of 10M English sentences [ 96 ], BERT’s performance ( F 1 m  = 81.8) was lower than RoBERTa. We noticed a performance drop for all models when comparing test-rw and test-mt . It seems that RoBERTa is better able to resolve in-sentence relations between multiple targets (performance degeneration of only up to − 0.6pp.) than BERT (− 2.9pp.). We suggest to use RoBERTa for TSC on news, since fine-tuning it was faster than fine-tuning XLNET and RoBERTa achieved similar or better performance than other LMs.

While the proposed model yielded competitive results on previous TSC datasets (Table 5.9 ), LCF was the top performing model. Footnote 16 When comparing the performances across all four datasets, the importance of the consolidation became apparent, e.g., performance was lowest on the Twitter set, where a simplistic consolidation was employed during the dataset’s creation (Sect. 5.3.3.4 ). The performance differences of individual models when contrasting their use on prior datasets and NewsMTSC highlight the need LCF performed consistently best on prior datasets but worse than the proposed model on NewsMTSC. One reason might be that LCF’s weighting approach relies on a static distance parameter, which seems to degrade performance when used on longer texts as in NewsMTSC (Sect. 5.3.3.6 ). When increasing LCF’s window width SRD, we noticed a slight improvement of 1pp. (SRD =  5) but degradation for larger SRD.

Ablation Study

We performed an ablation study to test the impact of four key factors: target mask, EKS, coreferential mentions, and fine-tuning the LM’s parameters. We tested all LMs and if not noted otherwise report results for RoBERTa since it generally performed best (Sect. 5.3.5 ). We report results for test-mt (performance influence was similar on either test set, with performances generally being ≈3–5pp. higher on test-rw ). Overall, we found that our changes to the initial design [ 151 ] contributed to an improvement of approximately 1.9pp. The most influential changes were the selected EKS and in part the use of coreferential mentions. Using the target mask input channel without coreferences and LM fine-tuning yielded insignificant improvements of up to 0.3pp. each. We did not test the VADER-based sentence classification proposed by Hosseinia, Dragut, and Mukherjee [ 151 ] since we expected no improvement by using it for various reasons. For example, VADER uses a dictionary created for a domain other than news and classifies the sentence’s overall sentiment and thus is target-independent.

Table 5.10 details the results of exemplary EKS, showing that the best combination (SENT, MPQA, and NRC) yielded an improvement of 2.6pp. compared to not using an EKS (zeros). The single best EKS (LIWC or SENT) each yielded an improvement of 2.4pp. The two EKS “no EKS” and “zeros” represented a model lacking the EKS input channel and an EKS that only yields 0’s, respectively.

The use of coreferences had a mixed influence on performance (Table 5.11 ). While using coreferences had no or even a negative effect in our model for large LMs (RoBERTa and XLNET), it can be beneficial for smaller LMs (BERT) or batch sizes (8). When using the modes “ignore,” “add coref. to mask,” and “add coref. as example,” we ignored coreferences, added them to the target mask, and created an additional example for each, respectively. Mode “none” represents a model that lacks the target mask input channel.

5.3.6 Error Analysis

To understand the limitations of the proposed model, we carried out a manual error analysis by investigating a random sample of 50 incorrectly predicted examples for each of the test sets. For test-rw , we found the following potential causes (not mutually exclusive): edge cases with very weak, indirect, or in part subjective sentiment (22%) or where both the predicted and true sentiment can actually be considered correct (10%) and sentiment of given target confused with different target (14%). The latter occurred especially often for long sentences consisting mostly of phrases that indicate the predicted sentiment but concerning a different target, e.g., “By the time he and Mr. Smith (predicted: negative, true: neutral) were trading texts, […], John was already fired by his boss.” Further, sentence’s sentiment was unclear due to missing context (10%) and the consolidated answer in NewsMTSC was wrong (10%). In 16%, we found no apparent reason. For test-mt , potential causes occurred approximately similarly often as in test-rw , except that targets are confused more often (20%).

5.3.7 Future Work

We identify three main areas for future work. The first area is related to the dataset. Instead of consolidating multiple annotators’ answers during the dataset creation, we propose to test to integrate the label selection into the model [ 295 ]. Integrating the label selection into the machine learning part could improve the classification performance. It could also allow us to include more sentences in the dataset, especially the edge cases that our restrictive consolidation currently discards.

To improve the model design, we propose to design the model specifically for sentences with multiple targets, for example, by classifying multiple targets in a sentence simultaneously. While we early tested various such designs, we did not report them due to their comparably poor performances. Further work in this direction should perhaps also focus on devising specialized loss functions that set multiple targets and their polarity into relation. Lastly, one can improve various technical details of the proposed model, e.g., by testing other interaction layers, such as LSTMs, or using layer-specific learning rates in the overall model, which can increase performance [ 347 ].

5.3.8 Conclusion

In this section, we presented NewsMTSC, a dataset for target-dependent sentiment classification (TSC) on news articles consisting of 11.3k manually annotated examples. Compared to prior TSC datasets, the dataset is different in key factors, such as that its texts are on average 50% longer, sentiment is expressed explicitly only rarely, and there is a separate test set for sentences containing multiple targets. In part as a consequence of these differences, state-of-the-art TSC models yielded non-optimal performances in our evaluation.

We proposed a model that uses a bidirectional GRU on top of a language model (LM) and other embeddings, instead of masking or weighting mechanisms as employed by the prior state of the art. We found that the proposed model achieved superior performances on NewsMTSC and was competitive on prior TSC datasets. RoBERTa yielded better results compared to using BERT, because RoBERTa is pre-trained on news and we found it can better resolve in-sentence relations of targets, i.e., RoBERTa can better distinguish the individual sentiments if multiple targets are present in a sentence.

In the context of the PFA approach, TSC represents a method which we propose to use in the target concept analysis component. Conceptually, the TSC method is simpler compared to the fine-grained framing effect classification proposed in Sect. 5.2 . However, at the same time, the TSC method represents a pragmatic alternative to imitating part of the manual frame analysis as conducted in social science research on media bias. Due to its simplicity, the TSC method achieves strongly higher classification performance than the approach for frame property identification.

5.4 Summary of the Chapter

This chapter presented frame analysis as the second and last analysis component of person-oriented framing analysis (PFA). The component aims to identify how persons are portrayed in the given news articles, both at the article and sentence levels. This task is difficult for various reasons, such as news articles rather implicitly or indirectly frame persons, for example, by describing actions performed by a person. In sum, reliably inferring how news articles and sentences portray persons is much more complex compared to prior work in related fields. For example, target-dependent sentiment classification (TSC) is concerned with inferring a sentence’s sentiment toward a target concept. TSC methods achieve high classification performance, but only on domains where authors explicitly state their attitude toward the targets, such as product reviews or Twitter posts. Because of this difficulty and the other issues highlighted in the chapter, prior approaches concerning frame analysis yield inconclusive or superficial results or require exacting manual effort. Thus, automatically and reliably identifying how persons are portrayed in news articles is essential for the success of PFA. We explored two approaches to enable target concept analysis: event extraction and coreference resolution.

Our first, exploratory approach identifies how a person is portrayed using so-called frame properties (Sect. 5.2 ). Frame properties are categories that represent predefined, topic-independent effects of political framing. As such, the approach aims to resemble how media bias is analyzed in social science research while avoiding the topic-specific and analysis question-specific frames used there. Early during our research on this approach, we conducted a short qualitative evaluation and found inconclusive results. The inconclusive results and the at that point already very high annotation cost are difficulties common among prior automated approaches that aim to resemble frame analyses (Sect. 5.2.5 ).

We took these issues as a motivation to explore a more pragmatic route to our frame analysis component. Specifically, we devised a dataset and deep learning model for target-dependent sentiment classification (TSC) in news articles (Sect. 5.3 ). Similar to the frame properties approach outlined previously, the TSC method aims to identify how a person is portrayed using categories representing predefined, topic-independent effects of political framing. In contrast to frame properties, TSC uses only a single dimension as a fundamental effect of framing: polarity, i.e., whether a person is portrayed positively, negatively, or neither. This way, we avoid the infeasibly high annotation cost and ambiguity of analyzing frames or frame derivatives while still capturing an essential framing effect. In contrast to any prior work, our method is the first to reliably classify sentiment in news articles ( F 1 m  = 83.1) despite the high level of interpretation required.

In the evaluation described in Chap. 6 , we will investigate whether analyzing only a single framing effect dimension, i.e., sentiment polarity, instead of fine-grained framing effects, such as the frame properties, suffices to identify meaningful person-oriented frames.

The work described in this section was jointly conducted with our research for the method for context-driven cross-document coreference resolution. To avoid redundancy while improving readability, we refer to the respective sections in Chap. 4 and briefly summarize the work in the section at hand.

BBC, Breitbart, Chicago Tribune, CNN, Los Angeles Daily News, Fox News, HuffPost, Los Angeles Times, NBC, The New York Times, Reuters, USA Today, The Washington Post, and The Wall Street Journal.

To classify whether an article reports on politics, we use a DistilBERT-based [ 310 ] classifier with a single dense layer and softmax trained on the HuffPost [ 244 ] and BBC datasets [ 115 ]. During the subsequent manual annotation, coders discard remaining, non-political articles.

For this task, we use spaCy v2.1: https://spacy.io/usage/v2-1 .

We drew a random sample of 300 examples and concluded in a two-person discussion that the sentiment in 8 examples could be perceived differently.

Epochs ∈{3,  4}; batch size ∈{16,  32}; learning rate ∈{2 e  − 5,  3 e  − 5,  5 e  − 5}; label smoothing regularization (LSR) [ 353 ], 𝜖  ∈{0,  0.2 }; dropout rate, 0.1; \(\mathcal {L}_2\) regularization, λ  = 10 −5 . We used Adam optimization [ 181 ], Xavier uniform initialization [ 109 ], and cross-entropy loss [ 113 ]. Where multiple values for a hyperparameter are given, we tested all their combinations in an exhaustive search.

Each row in Table 5.5 shows the results of the hyperparameters that performed best on the validation set.

Each dataset roughly approximates the US media landscape, i.e., POLUSA by design [ 96 ], and BF18 because it was crawled from a news aggregator on a daily basis [ 49 ].

We employ spaCy 2.1 ( https://github.com/explosion/spaCy/releases/tag/v2.1.8 ) and neuralcoref 4.0 ( https://github.com/huggingface/neuralcoref/releases/tag/v4.0.0 ).

Determined by spaCy.

Determined by neuralcoref.

To think from the author’s perspective is not to be confused with the “author level” defined by Balahur et al. [ 16 ].

Prior to submitting a batch of examples to MTurk, we add 6% test examples with unambiguous sentiment, e.g., “Mr. Smith is a bad guy.”

For readability, we showcase inputs as used for BERT. We adapt inputs correspondingly for other LMs.

Epochs ∈{2,  3,  4}; batch size ∈{8,  16} (due to constrained resources not 32); learning rate ∈{2 × 10−5, 3 × 10−5, 5 × 10−5}; label smoothing regularization (LSR) [ 353 ], 𝜖  ∈{0,  0.2 }; dropout rate, 0.1; \(\mathcal {L}_2\) regularization, λ  = 1 × 10−5; SRD for LCF ∈{3,  4,  5}. We used Adam optimization [ 181 ], Xavier uniform initialization [ 109 ], and cross-entropy loss.

For previous models, Table 5.9 lists results reported by their authors. In our experiments, we found 0.4–1.8pp. lower performance compared to the reported results.

Eneko Agirre et al. “SemEval-2015 Task 2: Semantic Textual Similarity, English, Spanish and Pilot on Interpretability”. In: Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2015, pp. 252–263. doi : https://doi.org/10.18653/v1/S15-2045 . url : http://aclweb.org/anthology/S15-2045 .

Eneko Agirre et al. “SemEval-2016 Task 1: Semantic Textual Similarity, Monolingual and Cross-Lingual Evaluation”. In: Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2016, pp. 497–511. isbn : 978-1-941643-95-2. doi : https://doi.org/10.18653/v1/S16-1081 . url : http://aclweb.org/anthology/S16-1081 .

Phyllis F. Agran, Dawn N. Castillo, and Dianne G. Winn. “Limitations of data compiled from police reports on pediatric pedestrian and bicycle motor vehicle events”. In: Accident Analysis and Prevention 22.4 (1990), pp. 361–370. issn : 00014575. doi : https://doi.org/10.1016/0001-4575(90)90051-L .

Stefano Baccianella, Andrea Esuli, and Fabrizio Sebastiani. “SentiWordNet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining.” In: Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC’10) . Vol. 10. Valletta, Malta: European Language Resources Association (ELRA), 2010, pp. 2200–2204. url : https://www.aclweb.org/anthology/L10-1531/ .

Alexandra Balahur et al. “Sentiment analysis in the news”. In: Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC’10) . Valletta, Malta: European Language Resources Association (ELRA), 2010. url : https://arxiv.org/abs/1309.6202 .

Matthew A. Baum and Tim Groeling. “New Media and the Polarization of American Political Discourse”. In: Political Communication 25.4 (2008), pp. 345–365. doi : https://doi.org/10.1080/10584600802426965 .

Douglas Biber and Edward Finegan. “Styles of stance in English: Lexical and grammatical marking of evidentiality and affect”. In: Text - Interdisciplinary Journal for the Study of Discourse 9.1 (1989). issn : 0165-4888. doi : https://doi.org/10.1515/text.1.1989.9.1.93 . url : https://www.degruyter.com/doi/10.1515/text.1.1989.9.1.93 .

Ceren Budak, Sharad Goel, and Justin M. Rao. “Fair and balanced? Quantifying media bias through crowdsourced content analysis”. In: Public Opinion Quarterly 80.S1 (2016), pp. 250–271. doi : https://doi.org/10.1093/poq/nfw007 .

Dallas Card et al. “Analyzing Framing through the Casts of Characters in the News”. In: Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg, PA, USA: Association for Computational Linguistics , 2016, pp. 1410–1420. doi : https://doi.org/10.18653/v1/D16-1148 . url : http://aclweb.org/anthology/D16-1148 .

Dallas Card et al. “The Media Frames Corpus: Annotations of FramesAcross Issues”. In: Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 2: Short Papers) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2015, pp. 438–444. doi : https://doi.org/10.3115/v1/P15-2072 . url : http://aclweb.org/anthology/P15-2072 .

Wei-Fan Chen et al. “Learning to Flip the Bias of News Headlines”. In: Proceedings of the 11th International Conference on Natural Language Generation . Stroudsburg, PA, USA: Association for Computational Linguistics, 2018, pp. 79–88. doi : https://doi.org/10.18653/v1/W18-6509 . url : http://aclweb.org/anthology/W18-6509 .

Xiao Chen et al. “Aspect Sentiment Classification with Document-level Sentiment Preference Modeling”. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics . Stroudsburg, PA, USA: Association for Computational Linguistics, 2020, pp. 3667–3677. doi : https://doi.org/10.18653/v1/2020.acl-main.338 . url : https://www.aclweb.org/anthology/2020.aclmain.338 .

Kyunghyun Cho et al. “Learning Phrase Representations using RNN Encoder- Decoder for Statistical Machine Translation”. In: Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2014, pp. 1724–1734. doi : https://doi.org/10.3115/v1/D14-1179 . url : http://aclweb.org/anthology/D14-1179 .

Junyoung Chung et al. “Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling”. In: (Dec. 2014). arXiv: 1412.3555. url : http://arxiv.org/abs/1412.3555 .

Domenic V. Cicchetti and Alvan R. Feinstein. “High agreement but low kappa: II. Resolving the paradoxes”. In: Journal of Clinical Epidemiology 43.6 (Jan. 1990), pp. 551–558. issn : 08954356. doi : https://doi.org/10.1016/0895-4356(90)90159-M . url : https://linkinghub.elsevier.com/retrieve/pii/089543569090159M .

Stefano DellaVigna and Ethan Kaplan. The Fox News Effect: Media Bias and Voting. Tech. rep. 3. Cambridge, MA: National Bureau of Economic Research, Apr. 2006, pp. 1187–1234. doi : https://doi.org/10.3386/w12169 . url : http://www.nber.org/papers/w12169.pdf .

Jacob Devlin et al. “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding”. In: Proceedings of the 2019 Conference of the North . Stroudsburg, PA, USA: Association for Computational Linguistics, 2019, pp. 4171–4186. doi : https://doi.org/10.18653/v1/N19-1423 . arXiv: 1810.04805. url : http://aclweb.org/anthology/N19-1423 .

Chunning Du et al. “Adversarial and Domain-Aware BERT for Cross- Domain Sentiment Analysis”. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics . Stroudsburg, PA, USA: Association for Computational Linguistics, 2020, pp. 4019–4028. doi : https://doi.org/10.18653/v1/2020.acl-main.370 . url : https://www.aclweb.org/anthology/2020.aclmain.370 .

RobertMEntman. “Framing: Toward Clarification of a Fractured Paradigm”. In: Journal of Communication 43.4 (Dec. 1993), pp. 51–58. issn : 0021-9916. doi : https://doi.org/10.1111/j.1460-2466.1993.tb01304.x . url : https://academic.oup.com/joc/article/43/4/51-58/4160153 .

Robert M. Entman. “Framing Bias: Media in the Distribution of Power”. In: Journal of Communication 57.1 (Mar. 2007), pp. 163–173. issn : 00219916. doi : https://doi.org/10.1111/j.1460-2466.2006.00336.x . url : https://academic.oup.com/joc/article/57/1/163-173/4102665 .

Zhengjie Gao et al. “Target-Dependent Sentiment Classification With BERT”. In: IEEE Access 7 (2019), pp. 154290–154299. issn : 2169-3536. doi : https://doi.org/10.1109/ACCESS.2019.2946594 . url : https://ieeexplore.ieee.org/document/8864964/ .

Lukas Gebhard and Felix Hamborg. “The POLUSA Dataset: 0.9M Political News Articles Balanced by Time and Outlet Popularity”. In: Proceedings of theACM/IEEE Joint Conference on Digital Libraries in 2020 . NewYork, NY, USA: ACM, Aug. 2020, pp. 467–468. isbn : 9781450375856. doi : https://doi.org/10.1145/3383583.3398567 . url : https://dl.acm.org/doi/10.1145/3383583.3398567 .

Deepanway Ghosal et al. “KinGDOM: Knowledge-Guided DOMainAdaptation for Sentiment Analysis”. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics . Stroudsburg, PA, USA: Association for Computational Linguistics, 2020, pp. 3198–3210. doi : https://doi.org/10.18653/v1/2020.acl-main.292 . url : https://www.aclweb.org/anthology/2020.aclmain.292 .

Xavier Glorot and Yoshua Bengio. “Understanding the difficulty of training deep feedforward neural networks”. In: Journal of Machine Learning Research . 2010, pp. 249–256.

Google Scholar  

Namrata Godbole, Manja Srinivasaiah, and Steven Skiena. “Large-Scale Sentiment Analysis for News and Blogs”. In: Proceedings of the International Conference onWeblogs and Social Media (ICWSM) 7 (2007), pp. 219–222.

Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning . MIT Press, 2016. url : http://www.deeplearningbook.org .

Derek Greene and Pádraig Cunningham. “Practical solutions to the problem of diagonal dominance in kernel document clustering”. In: Proceedings of the 23rd international conference on Machine learning - ICML ’06 . New York, New York, USA: ACM Press, 2006, pp. 377–384. isbn : 1595933832. doi : https://doi.org/10.1145/1143844.1143892 . url : http://portal.acm.org/citation.cfm?doid=1143844.1143892 .

Gregory Grefenstette et al. “Coupling Niche Browsers and Affect Analysis for an Opinion Mining Application”. In: Coupling Approaches, Coupling Media and Coupling Languages for Information Retrieval . Vaucluse, France: Le Centre de Hautes Etudies Internationales D’Informatique Documentaire, 2004, pp. 186–194. url : https://dl.acm.org/doi/abs/10.5555/2816272.2816290 .

Tim Groseclose and Jeffrey Milyo. “A Measure of Media Bias”. In: The Quarterly Journal of Economics 120.4 (Nov. 2005), pp. 1191–1237. issn : 0033-5533. doi : https://doi.org/10.1162/003355305775097542 . url : http://dx.doi.org/10.1162/003355305775097542 .

Nicole Gruber and Alfred Jockisch. “Are GRU Cells More Specific and LSTM Cells More Sensitive in Motive Classification of Text?” In: Frontiers in Artificial Intelligence 3 (June 2020). issn : 2624-8212. doi : https://doi.org/10.3389/frai.2020.00040 . url : https://www.frontiersin.org/article/10.3389/frai.2020.00040/full .

David Heise. “The semantic differential and attitude research”. In: Attitude measurement 4 (1970), pp. 235–253.

Marjan Hosseinia, Eduard Dragut, and Arjun Mukherjee. “Stance Prediction for Contemporary Issues: Data and Experiments”. In: Proceedings of the Eighth International Workshop on Natural Language Processing for Social Media . Stroudsburg, PA, USA: Association for Computational Linguistics, 2020, pp. 32–40. doi : https://doi.org/10.18653/v1/2020.socialnlp-1.5 . url : https://www.aclweb.org/anthology/2020.socialnlp-1.5 .

Minqing Hu and Bing Liu. “Mining and summarizing customer reviews”. In: Proceedings of the 2004 ACM SIGKDD international conference on Knowledge discovery and data mining - KDD ’04 . New York, New York, USA: ACM Press, 2004, p. 168. doi : https://doi.org/10.1145/1014052.1014073 . url : http://portal.acm.org/citation.cfm?doid=1014052.1014073 .

Qingnan Jiang et al. “A Challenge Dataset and Effective Models for Aspect-Based Sentiment Analysis”. In: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2019, pp. 6279–6284. doi : https://doi.org/10.18653/v1/D19-1654 . url : https://www.aclweb.org/anthology/D19-1654 .

Daniel Kahneman and Amos Tversky. “Choices, values, and frames.” In: American Psychologist 39.4 (1984), pp. 341–350. issn : 0003-066X. doi : https://doi.org/10.1037/0003-066X.39.4.341 . url : http://content.apa.org/journals/amp/39/4/341 .

Johannes Kiesel et al. “SemEval-2019 Task 4: Hyperpartisan News Detection”. In: Proceedings of the 13th International Workshop on Semantic Evaluation . Stroudsburg, PA, USA: Association for Computational Linguistics, 2019, pp. 829–839. doi : https://doi.org/10.18653/v1/S19-2145 . url : https://www.aclweb.org/anthology/S19-2145 .

Sung-Hee Kim, Hyokun Yun, and Ji Soo Yi. “How to filter out random clickers in a crowdsourcing-based study?” In: Proceedings of the 2012 BELIV Workshop on Beyond Time and Errors - Novel Evaluation Methods for Visualization - BELIV ’12 . New York, New York, USA: ACM Press, 2012, pp. 1–7. isbn : 9781450317917. doi : https://doi.org/10.1145/2442576.2442591 . url : http://dl.acm.org/citation.cfm?doid=2442576.2442591 .

Diederik P. Kingma and Jimmy Ba. “Adam: A Method for Stochastic Optimization”. In: 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7–9, 2015 , Conference Track Proceedings. 2015. url : http://arxiv.org/abs/1412.6980 .

Svetlana Kiritchenko et al. “NRC-Canada-2014: Detecting Aspects and Sentiment in Customer Reviews”. In: Proceedings of the 8th InternationalWorkshop on Semantic Evaluation (SemEval 2014) . Dublin, Ireland: Association for Computational Linguistics, 2014, pp. 437–442. doi : https://doi.org/10.3115/v1/s14-2076 .

Haewoon Kwak et al. “FrameAxis: Characterizing Framing Bias and Intensity with Word Embedding”. In: (Feb. 2020), pp. 1–24. arXiv: 2002.08608. url: http://arxiv.org/abs/2002.08608 .

Xin Li et al. “A Unified Model for Opinion Target Extraction and Target Sentiment Prediction”. In: Proceedings of the AAAI Conference on Artificial Intelligence 33 (July 2019), pp. 6714–6721. issn : 2374-3468. doi : https://doi.org/10.1609/aaai.v33i01.33016714 . url : https://aaai.org/ojs/index.php/AAAI/article/view/4643 .

Sora Lim et al. “Annotating and Analyzing Biased Sentences in News Articles using Crowdsourcing”. In: Proceedings of the 12th Language Resources and Evaluation Conference . Marseille, France: European Language Resources Association, 2020, pp. 1478–1484. url : https://www.aclweb.org/anthology/2020.lrec-1.184 .

Bing Liu. “Sentiment Analysis and Opinion Mining”. In: Synthesis Lectures on Human Language Technologies 5.1 (May 2012), pp. 1–167. issn : 1947-4040. doi : https://doi.org/10.2200/S00416ED1V01Y201204HLT016 . url : http://www.morganclaypool.com/doi/abs/10.2200/S00416ED1V01Y201204HLT016 .

Pengfei Liu, Shafiq Joty, and Helen Meng. “Fine-grained Opinion Mining with Recurrent Neural Networks and Word Embeddings”. In: Proceedings of the 2015 Conference on Empirical Methods inNatural Language Processing . Stroudsburg, PA, USA: Association for Computational Linguistics, 2015, pp. 1433–1443. doi : https://doi.org/10.18653/v1/D15-1168 . url : http://aclweb.org/anthology/D15-1168 .

Yinhan Liu et al. “RoBERTa: A Robustly Optimized BERT Pretraining Approach”. In: (July 2019). arXiv: 1907.11692. url : http://arxiv.org/abs/1907.11692 .

Merriam-Webster Inc. The Merriam-Webster DictionaryNew Edition . Springfield, MA, US, 2016, pp. 1–960. isbn : 978-0877792956.

Rishabh Misra. News Category Dataset. 2018. url : https://www.kaggle.com/rmisra/news-category-dataset (visited on 02/19/2021).

Saif Mohammad and Peter Turney. “Emotions Evoked by Common Words and Phrases: Using Mechanical Turk to Create an Emotion Lexicon”. In: Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text . Los Angeles, CA: Association for Computational Linguistics, 2010, pp. 26–34.

Sebastian Nagel. Common Crawl: News Crawl. 2016. url : https://web.archive.org/web/20191118111519/https://commoncrawl.org/2016/10/news-dataset-available/ (visited on 03/24/2021).

Preslav Nakov et al. “SemEval-2013 Task 2: Sentiment Analysis in Twitter”. In: Second Joint Conference on Lexical and Computational Semantics (SEM), Volume 2: Proceedings of the Seventh International Workshop on Semantic Evaluation (SemEval 2013) . Atlanta, GA, USA: Association for Computational Linguistics, 2013, pp. 312–320. url : https://www.aclweb.org/anthology/S13-2052/ .

Preslav Nakov et al. “SemEval-2016 Task 4: Sentiment Analysis in Twitter”. In: Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016) . San Diego, CA, USA: Association for Computational Linguistics, 2016, pp. 1–18. doi : https://doi.org/10.18653/v1/S16-1001 .

David Niven. Tilt? The search for media bias . Praeger, 2002. isbn : 978–0275975777.

Jeppe Norregaard, Benjamin Horne, and Sibel Adali. “NELA-GT-2018: A large multi-labelled news dataset for the study of misinformation in news articles”. In: Proceedings of the Thirteenth International AAAI Conference on Web and Social Media . AAAI Press, 2019, pp. 630–638.

Bo Pang and Lillian Lee. “Seeing stars”. In: Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics - ACL ’05 . Morristown, NJ, USA: Association for Computational Linguistics, 2005, pp. 115–124. doi : https://doi.org/10.3115/1219840.1219855 . url : http://portal.acm.org/citation.cfm?doid=1219840.1219855 .

Zizi Papacharissi and Maria de Fatima Oliveira. “News Frames Terrorism: A Comparative Analysis of Frames Employed in Terrorism Coverage in U.S. and U.K. Newspapers”. In: The International Journal of Press/Politics 13.1 (Jan. 2008), pp. 52–74. issn : 1940-1612. doi : https://doi.org/10.1177/1940161207312676 . url: http://journals.sagepub.com/doi/10.1177/1940161207312676 .

Souneil Park et al. “NewsCube”. In: Proceedings of the 27th international conference on Human factors in computing systems - CHI 09 . New York, New York, USA: ACM Press, 2009, p. 443. isbn : 9781605582467. doi : https://doi.org/10.1145/1518701.1518772 . url : http://dl.acm.org/citation.cfm?doid=1518701.1518772 .

Minh Hieu Phan and Philip O. Ogunbona. “Modelling Context and Syntactical Features for Aspect-based Sentiment Analysis”. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA, USA: Association for Computational Linguistics, 2020 , pp. 3211–3220. doi : https://doi.org/10.18653/v1/2020.acl-main.293 . url : https://www.aclweb.org/anthology/2020.acl-main.293 .

Maria Pontiki et al. “SemEval-2014 Task 4: Aspect Based Sentiment Analysis”. In: Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2014, pp. 27–35. doi : https://doi.org/10.3115/v1/S14-2004 . url : http://aclweb.org/anthology/S14-2004 .

Maria Pontiki et al. “SemEval-2015 Task 12: Aspect Based Sentiment Analysis”. In: Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015) . Stroudsburg, PA, USA: Association for Computational Linguistics, 2015, pp. 486–495. doi : https://doi.org/10.18653/v1/S15-2082 . url : http://aclweb.org/anthology/S15-2082 .

James Pustejovsky and Amber Stubbs. Natural Language Annotation for Machine Learning: A guide to corpus-building for applications . 1st ed. Sebastopol, CA, US: O’Reilly Media, Inc., 2012. isbn : 9781449306663.

Joaquin Quionero-Candela et al. Dataset Shift in Machine Learning . The MIT Press, 2009. isbn : 9780262170055.

Vikas C. Raykar et al. “Learning From Crowds”. In: The Journal of Machine Learning Research 11 (2010), pp. 1297–1322. doi : https://doi.org/10.5555/1756006.1859894 .

Marta Recasens, Cristian Danescu-Niculescu-Mizil, and Dan Jurafsky. “Linguistic Models for Analyzing and Detecting Biased Language”. In: Proceedings of the 51st Annual Meeting on Association for Computational Linguistics . Sofia, BG: Association for Computational Linguistics, 2013, pp. 1650–1659. isbn : 9781937284503. url : https://www.aclweb.org/anthology/P13-1162.pdf .

Alexander Rietzler et al. “Adapt or Get Left Behind: Domain Adaptation through BERT Language Model Finetuning for Aspect-Target Sentiment Classification”. In: Proceedings of the 12th Language Resources and Evaluation Conference . European Language Resources Association, 2020. url : https://www.aclweb.org/anthology/2020.lrec-1.607 .

JakobRogstadius et al. “An Assessment of Intrinsic and Extrinsic Motivation on Task Performance in Crowdsourcing Markets”. In: Proceedings of the Fifth International Conference on Weblogs and Social Media . Ed. by Lada Adamic, Ricardo Baeza-Yates, and Scott Counts. Barcelona, Spain: AAAI Press, 2011, pp. 1–9. url : https://www.aaai.org/ocs/index.php/ICWSM/ICWSM11/paper/view/2778 .

Sara Rosenthal, Noura Farra, and Preslav Nakov. “SemEval-2017 Task 4: Sentiment Analysis in Twitter”. In: Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017) . Vancouver, Canada: Association for Computational Linguistics, 2017, pp. 502–518. doi : https://doi.org/10.18653/v1/s17-2088 .

Victor Sanh et al. “DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter”. In: arXiv preprint arXiv: 1910.01108 (Oct. 2019). arXiv: 1910.01108. url : http://arxiv.org/abs/1910.01108 .

Aaron D. Shaw, John J. Horton, and Daniel L. Chen. “Designing incentives for inexpert human raters”. In: Proceedings of the ACM 2011 conference on Computer supported cooperative work - CSCW ’11. New York, New York, USA: ACM Press, 2011, p. 275. isbn : 9781450305563. doi : https://doi.org/10.1145/1958824.1958865 . url : http://portal.acm.org/citation.cfm?doid=1958824.1958865 .

Youwei Song et al. “Targeted Sentiment Classification with Attentional Encoder Network”. In: Artificial Neural Networks and Machine Learning - ICANN 2019: Text and Time Series . Cham, US: Springer International Publishing, 2019, pp. 93–103. doi : https://doi.org/10.1007/978-3-030-30490-4_9 . url : http://link.springer.com/10.1007/978-3-030-30490-4_9 .

Robert Speer and Catherine Havasi. “Representing General Relational Knowledge in ConceptNet 5”. In: Proceedings of the Eight International Conference on Language Resources and Evaluation (LREC’12) . Istanbul, Turkey: European Language Resources Association (ELRA), 2012. isbn : 978-2-9517408-7-7. url : http://www.lrec-conf.org/proceedings/lrec2012/pdf/1072_Paper.pdf .

Ralf Steinberger et al. “Large-scale news entity sentiment analysis”. In: RANLP 2017 - Recent Advances in Natural Language Processing Meet Deep Learning . Incoma Ltd. Shoumen, Bulgaria, Nov. 2017, pp. 707–715. isbn: 9789544520496. doi : https://doi.org/10.26615/978-954-452-049-6_091 . url : http://www.acl-bg.org/proceedings/2017/RANLP%202017/pdf/RANLP091.pdf .

Chi Sun, Luyao Huang, and Xipeng Qiu. “Utilizing BERT for Aspect-Based Sentiment Analysisvia Constructing Auxiliary Sentence”. In: Proceedings of the 2019 Conference of the North . Stroudsburg, PA, USA: Association for Computational Linguistics, 2019, pp. 380–385. doi : https://doi.org/10.18653/v1/N19-1035 . url : http://aclweb.org/anthology/N19-1035 .

Chi Sun et al. “How to Fine-Tune BERT for Text Classification?” In: Chinese Computational Linguistics . Cham, US: Springer International Publishing, 2019, pp. 194–206. doi : https://doi.org/10.1007/978-3-030-32381-3_16 . url : http://link.springer.com/10.1007/978-3-030-32381-3_16 .

Christian Szegedy et al. “Rethinking the Inception Architecture for Computer Vision”. In: Proceedings of the IEEE conference on computer vision and pattern recognition . IEEE, 2016, pp. 2818–2826. doi : https://doi.org/10.1109/CVPR.2016.308 . arXiv: 1512.00567. url : http://arxiv.org/abs/1512.00567 .

Amanda Taub. How Liberals Got Lost on the Story of Missing Children at the Border. 2018. url : https://web.archive.org/web/20191120151037/https://www.nytimes.com/2018/05/31/upshot/liberals-immigration-childrenborder-misinformation.html (visited on 02/18/2021).

Yla R. Tausczik and James W. Pennebaker. “The Psychological Meaning of Words: LIWC and Computerized Text Analysis Methods”. In: Journal of Language and Social Psychology 29.1 (Mar. 2010), pp. 24–54. issn : 0261-927X. doi : https://doi.org/10.1177/0261927X09351676 . url : http://journals.sagepub.com/doi/10.1177/0261927X09351676 .

Baldwin Van Gorp. “Where is the frame? Victims and intruders in theBelgian press coverage of the asylum issue”. In: European Journal of Communication 20.4 (2005), pp. 484–507. doi : https://doi.org/10.1177/0267323105058253 .

Alex Wang et al. “GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding”. In: Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP . Stroudsburg, PA, USA: Association for Computational Linguistics, 2018, pp. 353–355. doi : https://doi.org/10.18653/v1/W18-5446 . url : http://aclweb.org/anthology/W18-5446 .

Ralph Weischedel et al. OntoNotes 5.0. 2013. url : url https://web.archive.org/web/20190705173013 , https://catalog.ldc.upenn.edu/LDC2013T19 (visited on 02/19/2021).

Theresa Wilson, Janyce Wiebe, and Paul Hoffmann. “Recognizing contextual polarity in phrase-level sentiment analysis”. In: Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing - HLT ’05. Morristown, NJ, USA: Association for Computational Linguistics, 2005, pp. 347–354. doi : https://doi.org/10.3115/1220575.1220619 . url : http://portal.acm.org/citation.cfm?doid=1220575.1220619 .

Yonghui Wu et al. “Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation”. In: Preprint (Sept. 2016). arXiv: 1609.08144. url : http://arxiv.org/abs/1609.08144 .

Heng Yang. LC-ABSA. 2020. url : https://github.com/yangheng95/LCABSA/blob/master/models/lc_apc/lcf_bert.py (visited on 04/15/2021).

Da Yin, Tao Meng, and Kai-Wei Chang. “SentiBERT: A Transferable Transformer-Based Architecture for Compositional Sentiment Semantics”. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics . Stroudsburg, PA, USA: Association for Computational Linguistics, 2020, pp. 3695–3706. doi : https://doi.org/10.18653/v1/2020.acl-main.341 . url: https://www.aclweb.org/anthology/2020.acl-main.341 .

Biqing Zeng et al. “LCF: A Local Context Focus Mechanism for Aspect-Based Sentiment Classification”. In: Applied Sciences 9.16 (Aug. 2019), pp. 1–22. issn : 2076-3417. doi : https://doi.org/10.3390/app9163389 . url : https://www.mdpi.com/2076-3417/9/16/3389 .

Bowen Zhang et al. “Enhancing Cross-target Stance Detection with Transferable Semantic-Emotion Knowledge”. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics . Stroudsburg, PA, USA: Association for Computational Linguistics, 2020, pp. 3188–3197. doi : https://doi.org/10.18653/v1/2020.acl-main.291 . url : https://www.aclweb.org/anthology/2020.acl-main.291 .

Lei Zhang, ShuaiWang, and Bing Liu. “Deep learning for sentiment analysis: A survey”. In: Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 8.4 (July 2018). issn : 1942-4787. doi : https://doi.org/10.1002/widm.1253 . url : https://onlinelibrary.wiley.com/doi/abs/10.1002/widm.1253 .

Download references

Author information

Authors and affiliations.

Department of Computer Science, Humboldt University of Berlin, Berlin, Germany

Felix Hamborg

You can also search for this author in PubMed   Google Scholar

Rights and permissions

Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License ( http://creativecommons.org/licenses/by/4.0/ ), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

Reprints and permissions

Copyright information

© 2023 The Author(s)

About this chapter

Cite this chapter.

Hamborg, F. (2023). Frame Analysis. In: Revealing Media Bias in News Articles. Springer, Cham. https://doi.org/10.1007/978-3-031-17693-7_5

Download citation

DOI : https://doi.org/10.1007/978-3-031-17693-7_5

Published : 06 October 2022

Publisher Name : Springer, Cham

Print ISBN : 978-3-031-17692-0

Online ISBN : 978-3-031-17693-7

eBook Packages : Computer Science Computer Science (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

  • Publish with us

Policies and ethics

  • Find a journal
  • Track your research
  • Open access
  • Published: 12 February 2024

Experiences of delivering and receiving mental healthcare in the acute hospital setting: a qualitative study

  • Daniel Romeu   ORCID: orcid.org/0000-0002-2417-0202 1 , 2 ,
  • Elspeth Guthrie   ORCID: orcid.org/0000-0002-5834-6616 1 ,
  • Sonia Saraiva   ORCID: orcid.org/0000-0002-2305-9246 1 ,
  • Carolyn Czoski-Murray   ORCID: orcid.org/0000-0001-7742-2883 1 ,
  • Jenny Hewison   ORCID: orcid.org/0000-0003-3026-3250 1 &
  • Allan House   ORCID: orcid.org/0000-0001-8721-8026 1  

BMC Health Services Research volume  24 , Article number:  191 ( 2024 ) Cite this article

345 Accesses

Metrics details

Recent investment in UK liaison psychiatry services has focused on expanding provision for acute and emergency referrals. Little is known about the experiences of users and providers of these services. The aim of this study was to explore the experiences of users of acute liaison mental health services (LMHS) and those of NHS staff working within LMHS or referring to LMHS. A secondary aim was to explore the potential impact of a one-hour service access target on service delivery.

Cross-sectional qualitative study. Individual interviews were audio-recorded, transcribed verbatim and interpreted using framework analysis.

Service users reported mixed experiences of LMHS, with some reporting positive experiences and some reporting poor care. Most service users described the emergency department (ED) environment as extremely stressful and wished to be seen as quickly as possible. Staff described positive benefits of the one-hour access target but identified unintended consequences and trade-offs that affected other parts of the liaison service.

Conclusions

The assessment and treatment of people who attend ED with mental health problems needs to improve and particular attention should be given to the stressful nature of the ED environment for those who are extremely agitated or distressed.

Peer Review reports

The number of people attending emergency departments (EDs) in England has continued to rise, aside from the COVID-19 period. In 2019/20, there were 25.0 million ED attendances compared to 21.5 million in 2011/12 1 . In April 2022, waiting time performance in EDs was the worst recorded in modern data collections [ 1 ], and people with mental health problems had to wait substantially longer than those with physical health problems. Although mental health presentations decreased during lockdown, there was a bounce back post-lockdown with even greater numbers attending ED than before [ 2 ].

There are relatively few studies of people’s experiences of liaison mental health services (LMHS) in the UK. A recent internet survey of respondents’ experiences of LMHS in England showed that only 31% of service users found their contact with such services helpful [ 3 ]. Latent class analysis identified three types of experience; those who had a positive experience, those who reported a negative experience and those who were non-committal. Suggestions for improvement included the provision of a 24/7 service, reduced waiting times for assessment, and clearer communication about treatment or care post-assessment.

Prior studies of user satisfaction of LMHS in the UK have also been mixed [ 4 ]. One previous study which involved in-depth qualitative interviews with service users found that people complained about long waiting times before being able to access liaison services [ 5 ]. Some service users reported good experiences characterised by close collaboration between the service user and liaison practitioner whilst others described po or experiences.

In contrast with the UK, studies from Australia have reported positive service user experiences of LMHS with high levels of service user satisfaction [ 6 , 7 , 8 ]. In one study, service users reported timely access to being seen by a liaison practitioner and reported feeling listened to, understood and helped in a positive fashion, with an emphasis on problem solution [ 6 ].

All 170 hospitals in England with an ED now have at least a rudimentary LMHS [ 9 ]. These services have undergone substantial growth in the last seven years following significant investment from NHS England [ 10 ]. There has been particular expansion in acute services, and a “Core-24” service model has been developed, with staffing ratios based upon hospital size in terms of bed numbers [ 11 ]. These Core-24 teams usually consist of at least one liaison psychiatrist and several liaison mental health nurses. They focus on emergency work, providing 24-hour cover for EDs and acute ward referrals, with an emphasis on one-off assessments followed by signposting.

These new developments have been accompanied by rigorous performance targets for response times and throughput. In 2016, NHS guidance stated that a person experiencing a mental health crisis should receive a response from a LMHS within a maximum of 1 h of receipt of referral, and within 4 h the person should have received: “ a full biopsychosocial assessment if appropriate, and have an urgent and emergency mental health care plan in place, and as a minimum, be en route to their next location if geographically different, or have been accepted and scheduled for follow-up care by a responding service, or have been discharged because the crisis has resolved” [ 11 ]. Further review of access standards for mental health services in 2021 maintained the 1-hour target [ 9 ], although waiting time targets for all patients attending EDs are under review due to consistent and increasing failures to meet them.

While many hospitals have benefitted from the introduction of Core-24, especially where there were no or only rudimentary services previously, other established liaison services have had to change or modify their ways of working to meet targets. In addition to acute cover, these established services previously offered lower volume, higher intensity work involving the assessment, treatment and co-management of patients with complex physical and mental health problems seen in either inpatient or outpatient settings.

We previously completed interviews with 73 NHS staff from 11 hospital trusts in England who were either LMHS staff or worked closely with them and found that interviewees most valued being able to spend time with patients to carry out therapeutic interventions [ 12 ]. Some staff provided continued treatment for patients admitted to acute hospitals over several weeks. For example, in one service mental health nurses regularly visiting older adult patients or those on stroke wards to provide encouragement with eating and rehabilitation, both vital components of ensuring recovery. Teams with psychologists, therapists or mental health nurses trained in specific interventions (like cognitive behavioural therapy) offered brief interventions while the patient was admitted to an acute hospital bed, or a follow-up appointment after discharge. Staff reported problems with continuity of care across the secondary-primary interface; a lack of mental health resources in primary care to support discharge; a lack of shared information systems; a disproportionate length of time spent recording information instead of face-to-face patient contact; and a lack of a shared vision of care. Similar issues were identified across different liaison service types.

The aim of the present study was to better understand the experiences of users and providers of LMHS, and to explore hospital staff’s experiences of the changes brought about by the NHS England’s investment in Core-24 and any impact on patient care. We were particularly interested in improving our understanding of the mechanisms and trade-offs involved in relation to meeting one key performance target, the one-hour response time set by NHS England for LMHS. Recent programme theory suggests that the imposition of such fixed targets may have unintended consequences for liaison services and other parts of the health care system [ 13 ].

This work formed part of the first phase of a programme funded through the NIHR Health Services and Delivery Research scheme to evaluate the cost-effectiveness and efficiency of different configurations of liaison psychiatry services in England (LP-MAESTRO) [ 14 ]. The Consolidating Criteria for Reporting Qualitative Research (COREQ) guidelines [ 15 ] have been followed.

This was a cross-sectional qualitative study with service users of hospital-based LMHS and hospital staff with either experience of working in, or working closely with, LMHS.

Setting and sample

Service users were recruited from two Northern cities in England. We aimed to recruit 8–10 service user participants and developed a purposive sampling frame to ensure maximum variation. Potential participants were approached by either LMHS staff to determine their interest, or by local service user organisations who were invited to identify participants for the project through their own contacts. Once consent to contact had been provided by the service users, they were contacted directly by a member of the research team, who explained the study and provided a study information sheet. The potential participant was given at least 48 h to decide whether to participate. All participants provided written informed consent and there were no dropouts. No relationship was established with participants prior to study commencement. Participants were not informed of any of the interviewers’ personal goals for conducting the research.

Hospital staff were recruited from two hospitals in Northern England, both with EDs and within the same city. A maximum-divergence sampling frame was developed to maximise diversity according to professional background, sub-specialism within the LMHS, clinical or managerial focus and whether liaison team member or referrer to the service. Overall, we planned to recruit 8–10 staff participants. All staff participants provided written informed consent and there were no dropouts.

Data collection

Service users.

Nine service users were individually interviewed using a semi-structured topic guide. The service user topic guide was developed for this study (LP MAESTRO) and not published elsewhere (see Additional file 1 ). It consisted of a list of key topic areas with open-ended questions and additional prompts covering the following areas: introductory questions identifying the contact the participant had had with acute care; experiences of the acute care received from acute hospital staff; accounts of care received from LMHS staff; and views on desirable changes and ways to achieve them. They were not asked specifically about Core-24 developments, as it was unlikely that they would be familiar with such policy and staffing changes. However, staffing and waiting times were included as part of the topic guide.

Hospital Staff

Eight hospital staff were individually interviewed using a semi-structured topic guide. The hospital staff topic guide was adapted from an earlier topic guide used in the LP-MAESTRO study in relation to a previous investigation of liaison psychiatry and hospital staff experiences of liaison services [ 12 ]. The adapted topic guide, which focuses primarily on staff experiences of CORE-24 is provided in Additional file 2 . The following key topic areas were covered: introductory questions about the staff member’s work history and the nature of their involvement with LMHS; experiences of LMHS prior to introduction of Core-24; description of any changes resulting from Core-24; impact of these changes on the service; impact on patient care; and views on how the service could be improved.

Interviews lasted 30–90 min and took place via telephone between September 2017 and February 2019. Participant interviews were conducted first, followed by interviews with hospital staff. With permission all interviews were audio-recorded and transcribed verbatim. There were no repeat interviews. Transcripts were not returned to participants for comment or correction. Only the interviewer and participant were present at each interview. No field notes were recorded.

Interviews were conducted by three researchers, all from the Leeds Institute of Health Sciences and qualified by experience and training (CCG, SS, EG). None were involved in the delivery of acute LMHS at the time of the study. EG is a female Professor of Psychological Medicine and Consultant Psychiatrist. CCM is a female Senior Research Fellow and SS is a female Research Fellow. EG had previously worked in an acute liaison mental health team and was generally supportive of the Core-24 developments prior to the study. Neither SS nor CCM had a priori views or identified biases. The form and content of the topic guides were developed in collaboration with people with personal experience of mental health problems and accessing LMHS.

Data analysis

The semi-structured interviews were interpreted independently by DR and EG using framework analysis [ 16 ]. This is a qualitative method that is useful in research that has specific questions, a limited time frame and a pre-determined sample; it is therefore well-suited to applied policy research. First, DR and EG independently read all transcripts with the study’s aim in mind. Each then independently reviewed all transcripts line by line identifying relevant experiences, opinions, descriptions of incidents and emotions (codes). DR collated codes into a draft theoretical framework which was refined through discussion with EG. It became apparent to base several framework categories around the key areas of interest in the interview schedule as we wanted to be open to issues arising from the data. DR then matched the data to the provisional framework. Each example was independently included under one or more theme in the thematic chart by DR and EG who then met to resolve any disparities. In the final stage, findings were reviewed by AH. Relevant supporting quotations were then extracted from interview transcripts to illustrate each theme and sub-theme. Data from service users and staff were analysed separately but are presented together if relevant to the theme or sub-theme. Participants were not asked to provide feedback on the findings.

Sample characteristics

Seventeen in-depth interviews were conducted, nine with LMHS users and eight with healthcare professionals. Service user participants consisted of 3 men and 6 women with varying age ranges (Table  1 ). Presenting problems included self-harm, psychosis, mania, long-term physical health problems and medically unexplained symptoms. The interviewed professionals were mental health liaison nurses ( n  = 3), consultant liaison psychiatrists ( n  = 2), general nurses ( n  = 2) and one consultant in emergency medicine.

Main findings

Participants discussed a range of topics surrounding the provision and experience of mental healthcare in general hospitals. They illustrated the complexity involved in meeting mental health needs in this setting. Below we outline our findings in terms of themes and sub-themes that emerged from the interviews; the four themes and their constituent subthemes are summarised in Table  2 . The staff topic guide included specific questions about Core-24 that were not included in the service users’ topic guide, so most of the sub-themes around the Core-24 service standard are only relevant to staff participants.

Theme one: the emergency department (ED)

Healthcare professionals and service users discussed their views of the ED as a site for mental healthcare provision. The content of their discourse comprised the sub-themes of ED staff, physical environment, appropriateness for mental health problems and desired characteristics.

Service users recounted highly variable experiences of ED staff when seeking help for their mental health problems. Some were described as kind and compassionate people who acknowledged distress and evoked feelings of validation:

“ They recognised I wasn’t putting things on, that I did feel acutely suicidal as I was saying ” – Participant 1.

Others disclosed negative views of ED staff, describing them as unpleasant and harsh. Three participants reported that ED staff withheld treatment that they thought was needed. Others reported that staff did not allow the service user to speak and failed to provide any guidance or support on discharge.

“The GP referred me to A&E and when I arrived there, they were very very harsh” – Participant 6.

LMHS professionals generally had negative perceptions of ED staff, reporting that they had poor psychiatric knowledge and skills. Several felt that ED staff did not appreciate the role of LMHS and frequently made inappropriate referrals. Some suggested that ED staff had little interest in mental health problems:

“I think sometimes they don’t ask more questions about mental health, and I don’t know if that is because they don’t feel confident to, or they just don’t want to” – Participant 13.

Physical environment

The ED environment was discussed exclusively by service users, and their opinions were overwhelmingly negative. Common issues were that the assessment room was uncomfortable and small:

“You’re brought into this really small room with no windows, it was tiny, it was also not necessarily painted, it was very scruffy ” – Participant 5.

A lack of provision of refreshments contributed to a sense of discomfort. The privacy of the assessment environment varied; two individuals reported that they were assessed in a private space, and one was not:

“In the department with just a curtain pulled around, so it wasn’t very private” – Participant 1.

Appropriateness for mental health problems

Both service users and providers questioned the appropriateness of the ED for people with mental health needs. No participants felt that the ED was an appropriate place for these needs.

“A lot of them are quite vulnerable, and more at risk of accidental self-harm or sort of vulnerable from other people in the department and it’s A&E isn’t it, I wouldn’t it consider a very nice environment for people that are experiencing psychosis” – Participant 13.

Service users described their experiences of seeking care in the ED as anxiety-provoking and lonely. They acknowledged that attending the department was an undesirable last resort, only done when other services and professionals could not be accessed in the community.

“It’s not the best solution by any stretch of the imagination but, but it’s the only place that’s available” – Participant 8.

Desired characteristics

Participants suggested ways that the ED could be improved to better care for those with mental health needs. These included a more comfortable environment, the option to wait outside, better communication of next steps and knowledge of community-based support. One participant suggested the provision of company while awaiting input from the mental health team.

“I don’t know what else they could do apart from have somebody sit with you all the time until the psychiatrist came or somebody to assess you” – Participant 7.

Liaison practitioners also felt that the ED could be improved by providing more staff training in mental health assessments and improving referrals to the LMHS. This could reduce the volume of referrals and facilitate referral triage while reducing wait times for service users.

“If you upskill the ED people to even basic then liaison psychiatry should be able to turn down referrals… And we have to remember in the middle of all of this is a patient” – Participant 15.

Theme two: Liaison mental health services (LMHS)

The second theme refers to participants’ views and experiences of liaison mental health services (LMHS). There are three sub-themes: experiences of LMHS, barriers to contact and desired characteristics.

Service users described variable experiences of the help they received from LMHS. Contact with LMHS helped some individuals to feel more comfortable and to understand the next steps. Some described a therapeutic benefit of talking in depth about their issues:

“It helps me mental health, being able to talk about it and stuff” – Participant 3.

Others voiced that LMHS were either unhelpful or contributed to them feeling worse. This was related to the feeling of not being listened to and the perception that no tangible help or support was offered.

“ I’ve not got time for them as they do nothing for me” – Participant 9.

Some service users held the view that LMHS complete little more than a “box-ticking” exercise that offers little benefit to the service user. This was echoed by one of the physical healthcare professionals.

Common problems were that the professionals seemed rushed and incompetent. Three participants shared the view that LMHS staff were dismissive or disinterested. This led them to feel guilty and as though they had wasted the professional’s time.

“Sometimes the mental health staff can be very dismissive and treat me like I’ve just wasted everybody’s time, and I should have just looked after myself at home” – Participant 8.

Others described LMHS staff in a more positive light, reflecting that they allowed them to speak freely while listening carefully and acknowledging their needs. In some interviews, LMHS staff were described as caring and comforting. One participant felt that LMHS staff are underappreciated:

“I know with my experiences with liaison psychiatry that they do a lot more than people may think” – Participant 4.

Generally, interviewed professionals were complimentary towards LMHS staff, describing them as hard-working, knowledgeable, experienced, accessible, and committed to high-quality patient care. Participants had conflicting views on whether LMHS staff have a good relationship with the ward teams and whether they meet their expectations, although this was often attributed to a rise in demand for the service.

Barriers to contact

Participants discussed barriers to accessing LMHS. Some service users recounted how input from LMHS was postponed or withheld because they were under the care of a community mental health team. This sometimes resulted in interactions with a “diversion team”, which was described as a frustrating, obstructive experience.

“You just can’t get past diversion because they’ve been put in place to stop people like me who are known to the system… They’re basically there to go, ‘there, there, you’re ok, you go home and speak to your care coordinator tomorrow.’” – Participant 8.

Staff felt that significant barriers to contact with LMHS included insufficient staffing levels, particularly out of hours, and a seemingly excessive amount of time completing documentation.

“The [LMHS] team spend a long time writing things up and reporting… If we do make a referral for later in the evening or overnight, I don’t work nights, but they’re often told, ‘oh we can’t come and see the patient because we’re writing up our reports!’” – Participant 12.

Service users outlined factors that would improve their experience of receiving care from LMHS. Several participants described dissatisfaction with being discharged without a clear treatment plan and called for the provision of aftercare and more information about third sector organisations.

“If someone’s self-harming or whatever they shouldn’t just be discharged. They need aftercare and everything. It should be in their care plan.” – Participant 2.

Some described desirable characteristics of LMHS staff, which included compassion, knowledge, and clearer communication of delays and anticipated next steps. Service users expressed a desire to be treated as an individual and to be listened to attentively.

“You need front-line staff who have the personal interactive skills to acknowledge, to offer comfort and explain what is going to happen, not front-line staff who make you more agitated or that they are confused” – Participant 5.

Other desirable characteristics of the service identified include universal service provision across the country, a switch of focus from medications to psychosocial interventions, and a separate service for those who do not meet the criteria for admission but who feel unsafe to return home. Some service users voiced support for an acute mental health service separate from the ED.

Theme three: core-24 service standard

This theme encapsulates views towards the Core-24 service standard and the subthemes are the 1-hour wait, perceived benefits, unintended consequences , and policymaker detachment.

1-hour wait

Although professionals acknowledged the importance of targets, many felt that the one-hour target was unattainable, particularly for those with complex presentations or substance issues. Some felt that it was inappropriate to assume that service users’ needs are constant throughout the day. There was a consensus that immediacy was prioritised over clinical importance, which manifests as brief introductions within the hour instead of careful, comprehensive assessments.

“It’s not about how quickly you are seen, it’s about the quality of the interaction and I think if you are having to respond to patients in an hour that can sometimes compromise the quality” – Participant 16.

In contrast, service users almost universally expressed a wish to receive contact from the LMHS as soon as possible, and even a one hour wait felt too long to wait if someone was very distressed.

“When you are thinking of taking your own life, an hour is a lifetime” – Participant 1.

Perceived benefits (staff only)

The most salient benefit reported by the healthcare professionals was investment in the LMHS. They described more financial investment into the service, and the creation of staff posts to expand the workforce, contributing to feelings of reassurance and comfort. Although participants acknowledged the associated challenges of training new staff, overall, this change was perceived as positive.

“The investment within the services has enabled us to, erm you know, to broaden out what we do” – Participant 17.

Generally, professionals explained that the service standard improved patient flow. They felt that one-hour reviews were conducive to faster discharges and the prevention of unnecessary hospital admissions. They also reported a greater focus on the service user experience and acknowledged the target as an opportunity to improve the service further.

“I think that’s been a huge positive for the team because it’s made them think, actually, okay, we need to do this. How are we going to do it in the best way possible to get the service users experience and the standard of care for them as best as we can?” – Participant 14.

Unintended consequences (staff only)

Professionals also reported numerous undesired sequelae to the Core-24 one hour target. The first was that the target acted as an incentive for people to use the ED for their mental health needs in the knowledge that they would be seen quickly. This contributed to a rise in the clinical workload for both ED and LMHS staff.

“It was an odd thing to do when you’re trying to decrease attendances, it’s like a bit of an incentive to [attend the ED]” – Participant 11.

Some explained that the target had a detrimental impact on servicing providing ward cover, as LMHS staff are diverted from wards to the ED for initial reviews for new presentations. This results in delays on the wards and subsequently prolongs admissions.

“They used to see people who were in the beds before the parvolex (a treatment following a paracetamol overdose) ended, but they’re just unable to do that now because of the amount of people in A&E to be seen” – Participant 10.

The target has also had ramifications on working hours, with some participants reporting that their shifts were extended from eight to twelve hours, resulting in more lone working and reduced staff morale. This was identified as the reason for some staff members deciding to leave their jobs.

Policymaker detachment (staff only)

Generally, professionals felt that Core-24 was implemented poorly by policymakers and commissioners who were disconnected from the service. They described that no attempts were made to seek the views of clinicians, and that it was delivered as a compulsory change.

“The way that this change was brought in was very top-down, there was very little engagement with the team” – Participant 16.

One professional reported that they were informed with little notice that older people would be included in the remit of LMHS following the standard, and they received no formal training for this. The disconnect between policymakers and clinicians resulted in resentment among staff.

Service users echoed this idea by suggesting that policymakers were detached from the views and priorities of those seeking care. Some mentioned that these should be incorporated into decisions about LMHS provision:

“I think that the service should get more involvement from the service user’s experience” – Participant 6.

Theme four: stigma of mental illness

The final theme describes the stigma associated with mental health problems. The subthemes were discrimination and the mental-physical dichotomy .

Discrimination

Service users commonly felt discriminated against for having mental health problems. They described being treated differently to those with physical health problems, with their issues not being taken as seriously. Some recalled being dismissed and feeling guilty for accessing services.

“If you’re physically ill that counts, it’s given a higher priority over mental illness” – Participant 4.

Professionals also acknowledged the discrimination against those with mental health problems in the general hospital setting. They commented that service users with mental health needs are generally perceived as problematic and unwanted in the ED.

“Patients with mental health difficulties in the emergency department are the difficult ones, the bad ones, the ones that upset the data, or the ones that don’t move out quick enough” – Participant 15.

The mental-physical dichotomy

This subtheme describes the clear delineation between physical and mental health in the context of healthcare services. Both professionals and service users commented that mental health needs are frequently neglected in physical healthcare settings. This is attributed to a perceived unwillingness to enquire about psychiatric symptoms and a tendency to ignore biopsychosocial determinants of health.

“If I was to mention mental state, your consultants turn their faces away from me” – Participant 9.
“The traditional method of dealing with a lack of liaison psychiatry in the general hospital is to ignore the problem and just pretend it’s not there, to not notice that the patient is sad, not notice that they are anxious, to blame the patient, to discharge them early, to not take care of the wider side of psychosis difficulties that have prompted this admission” – Participant 15.

Clinicians also perceived a divide between mental and physical healthcare professionals. Some LMHS staff felt that ED clinicians had poor psychiatric knowledge and skills, that they often made inappropriate referrals with minimal information, and that their service was not understood or appreciated.

“I don’t think mental health is respected within the A&E department as a proper profession” – Participant 13.

Final analysis

The final stage of analysis is summarised in Table  3 , which shows comparisons across the service user and staff groups whilst also reflecting the strength of the signals from the data (determined by the proportion of participants who voiced these opinions). It shows some striking differences in patterns but also several areas of agreement. The one-hour access target is seen differently by service users and staff whilst issues related to stigma are perceived as important by both groups.

How our results compare

There are relatively few qualitative studies of LMHS, so this study is an important addition to the field. The variable experiences of LMHS users in this study are similar to those described by Eales and colleagues [ 5 ] and consistent with the recently published online survey of LMHS users [ 3 ]; some people reported good treatment and care from LMHS, whilst others report poor care and an unhelpful experience. Despite the increased funding for LMHS in recent years, people’s experiences remain patchy and well below the satisfaction levels reported by users of services in Australia [ 7 , 17 , 18 , 19 ]. However, it is difficult to compare services between countries with different healthcare systems.

Most service users felt that the ED environment contributed to additional stress and was an inappropriate place for people with acute mental health problems. This is consistent with previous studies [ 20 , 21 ], which have highlighted the negative and stressful aspects of the ED for people with mental health issues and described the ED as overstimulating and lacking in comfort and privacy. This is set against a backdrop of a recent survey carried out by the Royal College of Psychiatrists which reported that more than three quarters of people referred to mental health services resort to using emergency services because their mental health deteriorates whilst waiting for an initial assessment [ 22 ].

Wait-time targets

Opinions about the appropriateness and helpfulness of the one-hour performance target for LMHS varied between service users and staff. Service users highlighted the importance of being seen as quickly as possible in ED, particularly because the environment was stressful, but also because they were in a heightened state of distress and needed urgent relief. Some staff, however, believed the one-hour target distorted clinical practice with performance taking precedence over clinical need. This resulted in many unintended consequences including encouraging an increase in mental health ED attendances and a detrimental effect on other parts of the liaison service.

These findings support a logic model we previously developed to explain the impact an increase in liaison mental health provision may have on specific target response times [ 13 ]. Increased staffing levels initially enable LMHS to see more service users within the designated response target time, but various tensions and trade-offs within the system become apparent over time. If more service users attend ED due to the quicker response time, coupled with long waits in the community, pressure on the system increases again. This pressure causes a tension between the balance of ED work and the needs of patients with severe mental health problems who are inpatients in the acute hospital. The focus on ED and meeting the response target may result in potential disruptions to the care of hospital inpatients with deleterious clinical consequences and increased length of hospital stays. The introduction of a response target inevitably leads to unintended consequences in other parts of the healthcare system; the balance of advantages and disadvantages of the target across the whole system needs to be considered.

Public stigma and discrimination against people with mental illness is not a new phenomenon and is still widespread in society [ 23 ]. A review of 42 studies of ED staff attitudes towards service users presenting with mental health problems, 14 of which were conducted in the UK, reported widespread perceived negativity, although positive experiences were also acknowledged [ 24 ]. The findings from our study suggests negative attitudes towards people with mental health problems are still problematic in the ED setting. A recent qualitative systematic review exploring stigma and discrimination experienced by mentally ill individuals seeking care for physical and mental health concerns suggests that stigma and discrimination significantly compromise the quality of healthcare relationships with services users [ 25 ].

What can be done?

The Royal College of Emergence Medicine has produced a useful toolkit for improving care of people with mental health problems whilst in the ED, which stresses that all people with either a physical or mental health problem should have access to ED staff that understand and can address their condition [ 21 ]. There is a clear driver from both the Royal College of Emergency Medicine and the Royal College of Psychiatrists to improve the care of people with mental health problems who attend ED. There has also been recognition of this problem by NHS England with funding in 2017–2018 of £18 million for 234 winter mental health schemes to help alleviate pressures in ED for people with mental health problems [ 26 ]. Most of the funding was allocated to mental health liaison schemes, community crisis resolution and discharge and step-down schemes. Although many individual schemes reported local positive benefits, there were no robust evaluations which would support national rollout of any of these schemes.

There is some evidence that small positive attitude changes towards people with mental illness can be achieved by specific stigma reduction interventions [ 27 ], although relatively few interventions have been evaluated in the ED. Most educational interventions have focused solely on knowledge acquisition for specific conditions such as substance misuse disorders [ 28 ]. However, the endemic nature of stigma towards mental illness suggests that multi-level changes are required at organisational and personal levels. The Lancet Commission on ending discrimination in mental health included a review of all forms of stigma and discrimination against people with mental health conditions in all settings and societies globally [ 23 ]. The authors made several recommendations including policy and societal changes and workplace changes. Of relevance to the ED setting, they recommend that all healthcare staff receive mandatory training on the needs and rights of people with mental conditions, co-delivered by people with lived experience of mental health issues.

The staffing recommendations for Core-24 LMHS were largely based upon the size of hospital and the knowledge that mental health issues account for 4% of ED attendances. However, recent work suggests a further 4% of ED attendances consist of people attending with a physical health problem but who also have significant mental health issues [ 2 ]. This suggests that current LMHS staffing levels need to be reviewed, as Core-24 guidance may have underestimated the workload demands on LMHS, and workload is better estimated by patient throughput than the size of the hospital in terms of bed numbers or the presumed percentage of people in ED who may require liaison services [ 29 ].

Strengths and limitations

This study has several strengths. First, we met our recruitment targets, although, recruitment of service users took longer than we anticipated. There are no patient organisations that represent liaison service users, so recruitment can be challenging. However, we achieved a wide diversity of service user participants in terms of demographic characteristics and clinical problem areas. The most common clinical problems seen by liaison services in England are co-morbid physical and mental health problems, self-harm and cognitive problems [ 29 ]. Participants with co-morbid physical and mental health problems were represented in our participant sample. However, service users with cognitive problems were excluded from this study due to the inability to provide informed consent to participate. Second, the staff participants came from a range of professional backgrounds, including those who worked within LMHS and those who referred to LMHS. Third, we were able to explore both service user and staff perspectives about an important aspect of current service provision – the one-hour access target.

There were several limitations to the study. First, our sample size was relatively small and service user participants were only recruited from two geographical areas, and hospital staff from only two hospitals. We required members of staff who had experience of LMHS both prior to and subsequent to the introduction of Core-24, which limited the number of staff who we could interview and were willing to participate in the study. This also limited our ability to interview to the point of saturation. A larger staff sample may have resulted in other themes emerging so the findings of this study cannot therefore be generalised to other services in England, although many of the findings do accord with previous work in this area. Second, as discussed above, we were unable to recruit people with cognitive problems, making findings less relevant to liaison services for older adults. Third, interviews with participants and staff were conducted before the COVID-19 pandemic and its impact upon healthcare delivery. There was a marked drop-off in ED attendances during lockdown and the many liaison ED services were moved to other parts of the hospital to minimise spread of infection. Although there has been a clear bounce back in ED attendances among people with mental health problems post-pandemic [ 2 ], it is unclear to what extent services and service users have changed.

This study provides compelling evidence that the assessment and treatment of people who attend ED with mental health problems needs to further improve. The negative staff attitudes described are unacceptable, services for aftercare following assessment are inadequate, and the immediate experience in ED is often negative.

Particular attention should also be given to the stressful nature of the ED environment for those who are agitated or distressed. It can be argued that the ED is not the most appropriate place for people with acute mental health needs, but at present, there is often no clear alternative. Diversion schemes are under development in some areas. However, there will always be a need for many people with mental health problems to attend ED, as people with mental health issues commonly also have physical health problems, which require investigation and management in parallel with their mental health difficulties. Whilst ED service users emphatically support the one-hour response target, the imposition of such targets can have unintended consequences on other parts of the liaison service which need to be balanced to ensure parity for LMHS users in ED and those admitted in the acute hospital as inpatients.

Availability of data and materials

Data from this study are not available due to the qualitative nature of the study.

Abbreviations

Liaison Mental Health Service

Emergency Department

The King’s Fund. What’s going on with A&E waiting times? https://www.kingsfund.org.uk/projects/urgent-emergency-care/urgent-and-emergency-care-mythbusters(2022) . Accessed 10 Aug 2023.

Cann J, Barter R, Battle J, Schwenck J, Anakwe R. Patients presenting to an acute general hospital with acute mental health needs: a retrospective observational cohort study. BMJ Open. 2022;12(3):e059102.

Article   PubMed   PubMed Central   Google Scholar  

Guthrie E, Romeu D, Czoski-Murray C, Relton S, Walker A, Trigwell P, Hewison J, West R, Fossey M, Hulme C, House A. Experiences of people seen in an acute hospital setting by a liaison mental health service: responses from an online survey. BMC Health Serv Res. 2021;21(1):1–2.

Article   Google Scholar  

Eales S. Service users’ experiences of Liaison Mental Health Care (Doctoral dissertation, City University London). 2013.

Google Scholar  

Eales S, Callaghan P, Johnson B. Service users and other stakeholders’ evaluation of a liaison mental health service in an accident and emergency department and a general hospital setting. J Psychiatr Ment Health Nurs. 2006;13(1):70–7.

Article   PubMed   CAS   Google Scholar  

Wand T, White K, Patching J, Dixon J, Green T. An emergency department-based mental health nurse practitioner outpatient service: part 1, participant evaluation. Int J Ment Health Nurs. 2011;20(6):392–400.

Article   PubMed   Google Scholar  

Wand AP, Wood R, Fossey MJ, Aitken P. Standards, efficiency and effectiveness in consultation-liaison psychiatry. Aust NZJ Psychiatry. 2015;49(2):104–5.

Summers M, Happell B. Patient satisfaction with psychiatric services provided by a Melbourne tertiary hospital emergency department. J Psychiatr Ment Health Nurs. 2003;10(3):351–7.

NHS England. NHS England proposes new mental health access standards. 2021. https://www.england.nhs.uk/2021/07/nhs-england-proposes-new-mental-health-access-standards/ . Accessed 10 Aug 2023.

NHS England. Next steps on the NHS five year forward view. https://www.england.nhs.uk/wp-content/uploads/2017/03/NEXT-STEPS-ON-THE-NHS-FIVE-YEAR-FORWARD-VIEW.pdf . Accessed 10 Aug 2023.

NHS England. Achieving better access to 24/7 urgent and emergency mental health care. 2016. https://www.england.nhs.uk/wp-content/uploads/2016/11/lmhs-guidance.pdf . Accessed 10 Aug 2023.

Keeble J, Walker A, Guthrie E, Trigwell P, Quirk A, Hewison J, Murray CC, House A. Integrated liaison psychiatry services in England: a qualitative study of the views of liaison practitioners and acute hospital staffs from four distinctly different kinds of liaison service. BMC Health Serv Res. 2019;19(1):1–4.

House A, Guthrie E, Walker A, Hewsion J, Trigwell P, Brennan C, Crawford M, Murray CC, Fossey M, Hulme C, Martin A. A programme theory for liaison mental health services in England. BMC Health Serv Res. 2018;18:1–1.

NIHR. Liaison Psychiatry: Measurement And Evaluation of Service Types, Referral patterns and Outcomes LP-MAESTRO. (2015) https://fundingawards.nihr.ac.uk/award/13/58/08#/ (2015) . Accessed 10 Aug 2023.

Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups. Int J Qual Health Care. 2007;19(6):349–57.

Srivastava A, Thomson SB. Framework analysis: a qualitative methodology for applied policy research. 2009. JOAAG. 2022;4(2):72–9.

Wand T, White K, Patching J. Realistic evaluation of an emergency department-based mental health nurse practitioner outpatient service in Australia. Nurs Health Sci. 2011;13(2):199–206.

Wand T, White K, Patching J, Dixon J, Green T. Outcomes from the evaluation of an emergency department-based mental health nurse practitioner outpatient service in Australia. J Am Acad Nurse Pract. 2012;24(3):149–59.

Wand T, D’Abrew N, Barnett C, Acret L, White K. Evaluation of a nurse practitioner-led extended hours mental health liaison nurse service based in the emergency department. Aust Health Rev. 2014;39(1):1–8.

Schmidt M, Uman T. Experiences of acute care by persons with mental health problems: an integrative literature review. J Psychiatr Ment Health Nurs. 2020;27(6):789–806.

Swires-Hennessy K, Hayhurst C. Mental Health in Emergency departments: a toolkit for improving care. Online: The Royal College of Emergency Medicine; 2017.

Royal College of Psychiatrists. Hidden waits force more than three quarters of mental health patients to seek help from emergency services. 2022. https://www.rcpsych.ac.uk/news-and-features/latest-news/detail/2022/10/10/hidden-waits-force-more-than-three-quarters-of-mental-health-patients-to-seek-help-from-emergency-services . Accessed 10 Aug 2023.

Thornicroft G, Sunkel C, Aliev AA, Baker S, Brohan E, El Chammay R, Davies K, Demissie M, Duncan J, Fekadu W, Gronholm PC. The Lancet Commission on ending stigma and discrimination in mental health. Lancet. 2022;400(10361):1438–80.

Clarke D, Usick R, Sanderson A, Giles-Smith L, Baker J. Emergency department staff attitudes towards mental health consumers: a literature review and thematic content analysis. Int J Ment Health Nurs. 2014;23(3):273–84.

Tyerman J, Patovirta AL, Celestini A. How stigma and discrimination influences nursing care of persons diagnosed with mental illness: a systematic review. Issues Ment Health Nurs. 2021;42(2):153–63.

NHS England. Mental health and A&E pressures. 2018. https://www.england.nhs.uk/south-east/wp-content/uploads/sites/45/2018/11/Mental-health-and-ED-pressures-case-studies-of-schemes.pdf . Accessed 10 Aug 2023.

Na JJ, Park JL, LKhagva T, Mikami AY. The efficacy of interventions on cognitive, behavioral, and affective public stigma around mental illness: a systematic meta-analytic review. Stigma Health. 2022;7(2):127.

Gonzalez M, Clarke DE, Pereira A, Boyce-Gaudreau K, Waldman C, Demczuk L, Legare C. The impact of educational interventions on attitudes of emergency department staff towards patients with substance-related presentations: a quantitative systematic review. JBI Evid Synthesis. 2017;15(8):2153–81.

Saraiva S, Guthrie E, Walker A, Trigwell P, West R, Shuweidi F, Crawford M, Fossey M, Hewison J, Murray CC, Hulme C. The nature and activity of liaison mental services in acute hospital settings: a multi-site cross sectional study. BMC Health Serv Res. 2020;20(1):1–1.

General Assembly of the World Medical Association. World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. J Am Coll Dent. 2014;81(3):14–8.

Download references

Acknowledgements

Not applicable.

This project is funded by the National Institute for Health Research (NIHR) HS&DR programme (project reference 13/58/08). The work was also supported by a legacy provided by the family of Dr James Haigh. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care.

Author information

Authors and affiliations.

Leeds Institute of Health Sciences, University of Leeds, Leeds, UK

Daniel Romeu, Elspeth Guthrie, Sonia Saraiva, Carolyn Czoski-Murray, Jenny Hewison & Allan House

Leeds and York Partnership NHS Foundation Trust, Leeds, UK

Daniel Romeu

You can also search for this author in PubMed   Google Scholar

Contributions

AH, JH and EG conceived of the research. AH was the programme lead. CCM, SS and EG conducted the interviews. DR and EG conducted the analysis. DR and EG wrote the first draft of the manuscript. All authors (DR, EG, SS, CCM, JH, AH) contributed to the manuscript. All authors have read and approved the final manuscript.

Corresponding author

Correspondence to Daniel Romeu .

Ethics declarations

Ethics approval and consent to participate.

This research was conducted in accordance with the Declaration of Helsinki [ 30 ]. NHS ethical permission was obtained from the North of Scotland Research Ethics Service (REC reference: 15/NS/0025) and NHS Trust level approvals were obtained. All participants provided written informed consent to participate and there were no dropouts.

Consent for publication

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1., additional file 2., rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Romeu, D., Guthrie, E., Saraiva, S. et al. Experiences of delivering and receiving mental healthcare in the acute hospital setting: a qualitative study. BMC Health Serv Res 24 , 191 (2024). https://doi.org/10.1186/s12913-024-10662-4

Download citation

Received : 24 August 2023

Accepted : 30 January 2024

Published : 12 February 2024

DOI : https://doi.org/10.1186/s12913-024-10662-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

  • Liaison psychiatry
  • Mental health liaison services
  • Qualitative
  • Experiences of care

BMC Health Services Research

ISSN: 1472-6963

analysis frame research

  • Open access
  • Published: 09 February 2024

Key influences on university students’ physical activity: a systematic review using the Theoretical Domains Framework and the COM-B model of human behaviour

  • Catherine E. B. Brown 1 ,
  • Karyn Richardson 1 ,
  • Bengianni Halil-Pizzirani 1 ,
  • Lou Atkins 2 ,
  • Murat Yücel 3   na1 &
  • Rebecca A. Segrave 1   na1  

BMC Public Health volume  24 , Article number:  418 ( 2024 ) Cite this article

771 Accesses

Metrics details

Physical activity is important for all aspects of health, yet most university students are not active enough to reap these benefits. Understanding the factors that influence physical activity in the context of behaviour change theory is valuable to inform the development of effective evidence-based interventions to increase university students’ physical activity. The current systematic review a) identified barriers and facilitators to university students’ physical activity, b) mapped these factors to the Theoretical Domains Framework (TDF) and COM-B model, and c) ranked the relative importance of TDF domains.

Data synthesis included qualitative, quantitative, and mixed-methods research published between 01.01.2010—15.03.2023. Four databases (MEDLINE, PsycINFO, SPORTDiscus, and Scopus) were searched to identify publications on the barriers/facilitators to university students' physical activity. Data regarding study design and key findings (i.e., participant quotes, qualitative theme descriptions, and survey results) were extracted. Framework analysis was used to code barriers/facilitators to the TDF and COM-B model. Within each TDF domain, thematic analysis was used to group similar barriers/facilitators into descriptive theme labels. TDF domains were ranked by relative importance based on frequency, elaboration, and evidence of mixed barriers/facilitators.

Thirty-nine studies involving 17,771 participants met the inclusion criteria. Fifty-six barriers and facilitators mapping to twelve TDF domains and the COM-B model were identified as relevant to students’ physical activity. Three TDF domains, environmental context and resources (e.g., time constraints), social influences (e.g., exercising with others), and goals (e.g., prioritisation of physical activity) were judged to be of greatest relative importance (identified in > 50% of studies). TDF domains of lower relative importance were intentions, reinforcement, emotion, beliefs about consequences, knowledge, physical skills, beliefs about capabilities, cognitive and interpersonal skills, social/professional role and identity, and behavioural regulation. No barriers/facilitators relating to the TDF domains of memory, attention and decision process, or optimism were identified.

Conclusions

The current findings provide a foundation to enhance the development of theory and evidence informed interventions to support university students’ engagement in physical activity. Interventions that include a focus on the TDF domains 'environmental context and resources,' 'social influences,' and 'goals,' hold particular promise for promoting active student lifestyles.

Trial registration

Prospero ID—CRD42021242170.

Peer Review reports

Physical activity (PA) has a powerful positive impact on all aspects of health. Regular PA can prevent and treat noncommunicable diseases [ 1 , 2 ], build resilience against the development of mental illness [ 3 ], and attenuate cognitive decline [ 4 ]. Given these pervasive health benefits, increasing participation in PA is recognised as a global priority by international public health organisations. Indeed, a core aspect of the World Health Organisation’s action plan for a “healthier world” is to achieve a 15% reduction in the global prevalence of physical inactivity by 2030 [ 5 ].

Despite international efforts to reduce physical inactivity, university students frequently do not meet the recommended level of PA required to attain its health benefits. Approximately 40–50% of university students are physically inactive [ 6 ], many of whom attribute their inactivity to unique challenges associated with university life. For many students, the transition to university coincides with new academic, social, financial, and personal responsibilities [ 7 ], disrupting established routines and imposing additional barriers to the initiation or maintenance of healthy lifestyle habits such as regular PA [ 8 ]. Students’ PA tends to decline further during periods of high stress and academic pressure, such as exams and assignment deadlines [ 9 ]. This pattern has been observed across diverse university populations and cultural contexts [ 10 , 11 , 12 ], highlighting the importance of understanding the factors that contribute to physical inactivity among this cohort globally.

Understanding the barriers and facilitators to PA in the context of the university setting is an important step in developing effective, targeted interventions to promote active lifestyles among university students. A recently published systematic review found that lack of time, motivation, access to places to practice PA, and financial resources were primary barriers to PA for undergraduate university students [ 13 ]. A corresponding and complementary synthesis of the facilitators of PA, however, has not yet been conducted. Such a synthesis would be valuable in enabling a comprehensive understanding of the factors that influence students' PA and identifying facilitators that could be leveraged in intervention design. Furthermore, applying theoretical frameworks to understand barriers and facilitators to PA can guide the development of theory-informed, evidence-based interventions for university students that purposely and effectively target factors that influence their participation in PA.

The Theoretical Domains Framework (TDF) [ 14 , 15 , 16 ] and the COM-B model of behaviour [ 17 ] are two robust, gold-standard frameworks frequently used to examine the determinants of human behaviour. The TDF is an integrated framework of 14 theoretical domains (see Additional file 1 for domains, definitions, and constructs) which provide a comprehensive understanding of the key factors driving behaviour. The TDF was developed through expert consensus, synthesising 33 psychological theories (such as social cognitive theory [ 18 , 19 ] and the theory of planned behaviour [ 20 , 21 ] and 128 theoretical constructs (such as ‘competence’, ‘goal priority’, etc.) across disciplines identified as most relevant to the implementation of behaviour change interventions. Identifying the relative importance of theoretical domains allows intervention designers to triage which behaviour change strategies should be prioritised in intervention development [ 22 , 23 ]. The TDF has been widely applied by researchers and practitioners to systematically identify which theoretical domains are most relevant for understanding health behaviour change and policy implementation across a range of contexts, including education [ 24 ], healthcare [ 25 ], and workplace environments [ 26 ].

The 14 TDF domains map onto the COM-B model (Fig.  1 ), which is a broader framework for understanding behaviour and provides a direct link to intervention development frameworks. The COM-B model posits that no behaviour will occur without sufficient capability, opportunity, and motivation. Where any of these are lacking, they can be strategically targeted to support increased engagement in a desired behaviour, including participation in PA. Within the COM-B model, capability can be psychological (e.g., knowledge to engage in the necessary processes) or physical (e.g., physical skills); opportunity can be social (e.g., interpersonal influences) or physical (e.g., environmental resources); and motivation can be automatic (e.g., emotional reactions, habits) or reflective (e.g., intentions, beliefs). The COM-B model was developed through a process of theoretical analysis, empirical evidence, and expert consensus as a central part of a broader framework for developing behaviour change interventions known as the Behaviour Change Wheel (BCW) [ 17 ].

figure 1

The TDF domains linked to the COM-B model subcomponents

Note. Reproduced from Atkins, L., Francis, J., Islam, R., et al. (2017) A guide to using the Theoretical Domains Framework of behaviour change to investigate implementation problems. Implementation Science 12, 77.  https://doi.org/10.1186/s13012-017-0605-9

Using the TDF and COM-B model to understand the barriers and facilitators to university students’ participation in PA is valuable to inform the development of effective evidence-based interventions that are tailored to address the most influential determinants of behaviour change. As such, this systematic review aimed to: a) identify barriers and facilitators to university students’ participation in PA; b) map these factors using the TDF and COM-B model; and c) determine the relative importance of each TDF domain.

Study design

The systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [ 27 ]. The review protocol was registered on PROSPERO (CRD42021242170).

Search strategy

Search terms and parameters were developed in collaboration with a Monash University librarian with expertise in systematic review methodology. The following databases were searched on 15.03.2023 to identify relevant literature: MEDLINE, PsycINFO, and SPORTDiscus. Key articles were also selected for citation searching via Scopus. In consultation with a librarian, these databases were selected due to their unique scope, relevance, broad coverage, and utility. This process ensured the identified literature aligned with the aim and research topic of our systematic review. A 01.01.2010—15.03.2023 publication period was purposefully specified to account for the significant advancements in digital fitness support and tracking tools within the past decade [ 28 ], All available records were searched using the following combination of concepts in the title or abstract of the article: 1) barriers, facilitators, or intervention, Footnote 1 2) physical activity, 3) university, and 4) students. Each search concept was created by first developing a list of search terms relevant to each concept (e.g., for the ‘physical activity’ concept search terms included ‘physical exercise’, ‘physical fitness’, ‘sports’, ‘inactive’, ‘sedentary’, etc.). To create each concept, search terms were then searched collectively using the operator ‘OR’. Each search concept was then combined into the final search by using the operator ‘AND’. Search terms related to concepts 1, 2 and 3 included indexed terms unique and relevant to each database (i.e., Medical Subject Heading Terms for MEDLINE, Index Terms for PsycINFO, and Thesaurus terms for SPORTDiscus). The search was performed according to Boolean operators (e.g., AND, OR) (see Additional file 2 for the complete search syntax for MEDLINE). Unpublished studies were not sought.

Selection criteria

Articles were included if they: (a) reported university students’ self-reported barriers and/or facilitators to physical activity or exercise Footnote 2 ; (b) were written in English; and (c) were peer-reviewed journal articles. Articles encompassed studies directly investigating barriers and/or facilitators to students’ participation in PA and physical exercise intervention studies, where the latter reported participants’ self-reported barriers and/or facilitators to intervention adherence (see Table  1 below for full criteria).

Study selection

Identified articles were uploaded to EndNote X9 software [ 30 ]. A duplication detection tool was used to detect duplicates, which were then screened for accuracy by CB prior to removal. The remaining articles were uploaded to Covidence to enable blind screening and conflict resolution. Articles were screened at the title and abstract level against the inclusion and exclusion criteria by author CB, and 25% were independently screened by BP. The full text of studies meeting the inclusion criteria was then screened against the same criteria by CB, and 25% were again independently screened by BP. Differences were resolved by an independent author (KR). Inter-rater agreement in screening between CB and BP was high (0.96 for title and abstract screening, 0.83 for full-text screening). The decision to dual-screen 25% of studies was strategically chosen to balance thoroughness with efficiency, ensuring both the validity of the screening criteria and the reliability of the primary screener’s decisions. This approach aligns with the protocols used in similar systematic reviews in the field (e.g., [ 31 , 32 ]).

Data extraction

Key article characteristics were extracted, including the author/s, year of publication, country of origin, participant characteristics (e.g., enrolment status, exercise engagement [if reported]), sample size, research design, methods, and analytical approach. Barriers and facilitators were also extracted for each article and subsequently coded according to the 14 domains of the TDF and six subcomponents of the COM-B model. Quantitative data were only extracted if ≥ 50% of students endorsed a factor as a barrier or facilitator. This cut-off criterion was applied to maintain focus on the most common variables of influence and aligns with other reviews synthesising common barriers and facilitators to behaviour change (e.g., [ 26 , 33 ]).

A coding manual was developed to guide the process of mapping barriers and facilitators to the TDF and COM-B. All articles were independently coded by at least two authors (CB and BS, BP or KR). The first version of the manual was developed a priori, based on established guides for applying the TDF and COM-B model to investigate barriers and facilitators to behaviour [ 14 , 34 ], and updated as needed via regular consultation with a co-author and TDF/COM-B designer LA to ensure the accuracy of the data extraction. Barriers and facilitators were only coded to multiple TDF domains if deemed essential to accurately contextualise the core elements of the barrier/facilitator, and when the data in individual papers was described in sufficient detail to indicate that more than one domain was relevant. For example, if ‘lack of time due to competing priorities’ was reported as a barrier to PA, this encompassed both the ‘environmental context and resources’ (i.e., time) and ‘goals’ (i.e., competing priorities) domains of the TDF. Coding conflicts were resolved via discussion with LA.

Data analysis

The following three-step method was utilised to synthesise quantitative and qualitative data:

Framework analysis [ 35 ] was conducted to deductively code barriers and facilitators onto TDF domains and COM-B subcomponents. This involved identifying barriers and facilitators in each article, extracting and labelling them, and determining their relevance against the definitions of the TDF domains and COM-B subcomponents. This process involved creating tables to assist in the systematic categorisation of barriers and facilitators into relevant TDF domains and COM-B subcomponents.

Within each TDF domain, thematic analysis [ 36 ] was conducted to group similar barriers and facilitators together and inductively generate summary theme labels.

The relative importance of each TDF domain was calculated according to frequency (number of studies), elaboration (number of themes) and the identification of mixed barriers/facilitators regarding whether a theme was a barrier or facilitator within each domain (e.g., if some participants reported that receiving encouragement from their family to exercise was a facilitator, and others reported that lack of encouragement from their family to exercise was a barrier). The rank order was determined first by frequency, then elaboration, and finally by mixed barriers/facilitators.

This methodology follows previous studies using the TDF and COM-B to characterise barriers and facilitators to behaviour change and rank their relative importance [ 22 , 23 ].

Study characteristics

Following the removal of duplicates, 6,152 articles met the search criteria and were screened based on title and abstract. A total of 5,995 articles were excluded because they did not meet the inclusion criteria (see Fig.  2 below for the PRISMA flowchart). After the title and abstract screening, 157 full-text articles were retrieved and assessed for eligibility. One additional article was identified and included following citation searching of selected key articles. Thirty-nine articles met the inclusion criteria (see Additional file 3 for a summary of these studies). Eight studies were conducted in the USA, seven in Canada, three in Germany, two each in Qatar, Spain, the United Arab Emirates, and the United Kingdom, and one each in Australia, Belgium, Columbia, Egypt, Ireland, Japan, Kuwait, Malaysia, New Zealand, Saudi Arabia, South Africa, Sri Lanka, and Uganda.

figure 2

PRISMA flowchart illustrating the article selection process

Relative importance of TDF domains and COM-B components

Twelve of the 14 TDF domains and all six subcomponents of the COM-B model were identified as relevant to university students' PA. The rank order of relative importance of TDF domains and associated COM-B subcomponents are presented in Table  2 . The three most important domains were identified in at least 54% of studies.

Barriers and facilitators to student’s physical activity

Within the TDF domains, 56 total themes were identified, including 26 mixed barriers/facilitators, 18 facilitators and 12 barriers (Table  3 ). The barriers and facilitators identified within each TDF domain are summarised below (with associated COM-B subcomponent presented in parentheses), in order of relative importance:

1. Environmental context and resources (Physical Opportunity) ( n  = 90% studies)

The most frequent barrier to PA across all TDF domains was ‘lack of time’, most often in the context of study demands. Time constraints were exacerbated by long commutes to university, family responsibilities, involvement in co-curricular activities, and employment commitments. Students’ need for ‘easily accessible exercise options, facilities and equipment’ was a recurring theme. PA was deemed inaccessible if exercise facilities and other infrastructure to support PA, such as bike paths and running trails, were situated too far from the university campus or students’ residences, or if fitness classes were scheduled at inconvenient times. ‘Financial costs’ emerged as a theme. The costs associated with accessing exercise facilities, equipment and programs consistently deterred students from engaging in PA. The desire for ‘safe and enjoyable’, ‘weather appropriate’ environments for PA were frequently reported. Participating in outdoor PA in green spaces or near water increased enjoyment, provided the environment felt safe and weather conditions were suitable for PA. Factors related to students’ home, work, and university environment impacted their participation in ‘incidental PA’. Incidental PA was influenced by whether students engaged in domestic house chores, and manual work, and actively commuted to university and between classes on-campus. Students’ ‘access to a variety of physical activities’ and ‘information provision regarding on-campus exercise options’ impacted their PA. Students most often had access to a wide variety of physical activities, however, it could be difficult to access information about what types of activities were available on-campus and how to sign up to participate. The ‘lack of personalised physical activities to cater to individual fitness needs’ was a barrier, particularly for students with low levels of PA who required beginner-oriented programs. Another barrier was the ‘lack of university policy and promotion to encourage PA’, which led students to perceive that there was no obligation to participate in PA and that the university did not value it. ‘Health-concerning behaviours associated with university’, including poor diet, increased alcohol intake and sedentary behaviour, negatively impacted students’ PA. ‘Listening to music while exercising’ was a facilitator.

2. Social influences (Social Opportunity) ( n  = 72% studies)

Within social influences, ‘exercising with others’ emerged as the most frequent theme. Doing so increased students’ accountability, enjoyment and motivation, and helped them to overcome feelings of intimidation when exercising alone. Having a lack of friends to exercise with was a particular concern for students who were new to exercise or infrequently participated in PA. Receiving ‘encouragement from others to be physically active’, such as family members, friends, peers, and fitness instructors, shaped students’ values toward PA and enhanced their motivation and self-efficacy. Students’ family members, friends and teachers discouraged PA if it was not valued, or in favour of other priorities, such as academic commitments. Another recurrent theme was ‘competition or relative comparison to others’. While most students were motivated by competition, a minority felt demotivated if they compared themselves to others with higher PA standards, especially if they failed to achieve similar PA goals. Sociocultural norms influenced barriers/facilitators to PA across different cultures, and between various groups, such as international versus domestic students, and women versus men. Students from Japan and Hawaii viewed PA as an important part of their culture, in contrast to students from the Philippines who described the opposite. Participation in PA enabled international students to integrate with domestic students and learn about the local culture, however cultural segregation was a barrier to participation in university team sports. For female students from some middle-eastern countries, including Saudi Arabia, the UAE and Qatar, cultural norms made it impermissible for women to engage in PA, particularly compared to men. Religion also differentially impacted barriers/facilitators between women and men. Muslim women reported that Islamic practices, such as needing to engage in PA separately from men, be accompanied by a male family member while going outdoors, or dress modestly, posed additional barriers to PA. However, one study reported that Islamic teachings generally encouraged PA for both women and men by emphasising the importance of maintaining good health. Other gender-specific barriers were identified. Women often felt unwelcome or intimidated by men in exercise facilities, partly due to the perception that these facilities were tailored toward “masculine” sports and/or dominated by men. ‘Being stared at while engaging in PA’ was another barrier, impacting both women and students with a disability. A less common facilitator was the influence of both positive and negative ‘exercise role models’. For example, students practiced PA because they aspired to be like someone who was physically active, or because they did not want to be like someone who was not physically active.

3. Goals (Reflective Motivation) ( n  = 54%)

‘Prioritisation of PA compared to other activities’ was the most common theme within goals. Students frequently prioritised other activities, such as study, social activities, or work, over PA. However, those who played team sports or regularly practiced PA were more inclined to prioritise it for its recognised health benefits (i.e., stress management), and its role in enhancing confidence. Additional facilitators included ‘engaging in PA to achieve an external goal’, such as improving one’s appearance, and ‘setting specific PA-related goals’ as a means to enhance accountability.

4. Intentions (Reflective Motivation) ( n  = 44%)

Within intentions, ‘motivation to engage in PA’ was the most common theme. Students most often noted a lack of self-motivation for PA. Less frequent barriers included perceiving PA as an obligatory or necessary "chore", and ‘failing to follow through on intentions to engage in PA’. Conversely, ‘self-discipline to engage in PA’ emerged as a facilitator that assisted students in maintaining a regular PA routine.

5. Reinforcement (Automatic Motivation) ( n  = 38%)

The most frequent facilitator within reinforcement was ‘experiencing the positive effects of PA’ on their health and wellbeing. These included physical health benefits (i.e., maintaining fitness), psychological benefits (i.e., stress reduction), and cognitive health benefits (i.e., enhanced academic performance). Conversely, barriers arose from ‘experiencing discomfort during or after PA’ due to pain, muscle soreness or fatigue. ‘Past and current habits and routines’ was a theme. Students were more likely to participate in PA if they had established regular exercise routines, and that forming these habits at an early age made it easier to maintain them later in life. However, maintaining a regular PA routine was difficult in the context of inflexible university schedules. Students’ ‘sense of accomplishment in relation to PA’ was a theme. Students were less likely to feel a sense of accomplishment after participating in PA if it was not physically challenging. Consistent facilitators were ‘receiving positive feedback from others’ after engaging in PA, such as compliments, and ‘receiving incentives’, such as reducing the cost of gym memberships if students participated in more PA. ‘Experiencing a sense of achievement’ after reaching a PA-related goal or winning a sports match also served as a facilitator.

6. Emotion (Automatic Motivation) ( n  = 38%)

‘Enjoyment’ was the most frequently cited emotional theme. Most students reported that PA was fun and/or associated with positive feelings, however, a minority described PA as unenjoyable, boring, and repetitive. Students’ ‘poor mental health and negative affectivity’ (such as feeling sad, stressed or self-conscious, as well as fear of injury and pain), adversely impacted their motivation to be physically active.

7. Beliefs about consequences (Reflective Motivation) ( n  = 31%)

‘Beliefs about the physical health consequences of PA’ was the most recurrent barrier/facilitator. Most students understood that PA was essential for maintaining good health and preventing illness. However, some students who rarely or never engaged in PA believed they could delay pursuing an active lifestyle until they were older without compromising their health. Participating in PA to ‘maintain or improve one’s physical appearance’ acted as a facilitator. This motivation was most often cited in contexts such as increasing or decreasing weight, changing body shape or enhancing muscle tone. Beliefs about the positive environmental, occupational and psychological impacts of PA also served as facilitators. Students were motivated to participate in PA due to the environmental benefits of using active transport. They also acknowledged the importance of being physically fit for work and believed that being active was beneficial for mental health. ‘Receiving advice to participate in PA from a credible source’, such as a health professional, further facilitated students’ motivation to be active.

8. Knowledge (Psychological Capability) ( n  = 28%)

'Knowledge about the benefits of PA’, encompassing an understanding of the various types of benefits (i.e., physical, mental, or cognitive) and the biological mechanisms by which PA brings about these changes was identified as the most common knowledge theme. Being aware of these benefits positively influenced students’ motivation to be physically active. Conversely, students’ lack of knowledge about the gym environment and the programs available were barriers to PA. Regarding the gym environment, students’ ‘lack of knowledge about how to navigate through the gym, what exercises to do, and how to use exercise equipment’ amplified feelings of intimidation. Likewise, ‘lack of knowledge about the types of exercise programs and activities that were available on-campus, and how to sign up to participate’ were all barriers. A unique theme emerged concerning ‘knowledge about how to adapt physical activities for students with a disability’. Students with a disability described how fitness instructors often had a limited understanding of how to modify activities to enable them to participate. However, students with a disability were able to overcome this barrier if they possessed their own knowledge about how to tailor physical activities to meet their specific needs.

9. Physical skills (Physical Capability) ( n  = 21%)

The most prevalent theme within physical skills was ‘having the physical skills and fitness to participate in PA’. A lack of physical skills was most frequently a hindrance to PA. Additional obstacles to PA included being physically inhibited due to a ‘lack of energy’ or ‘physical injury’.

10. Beliefs about capabilities (Reflective Motivation) ( n  = 18%)

Within beliefs about capabilities, ‘self-efficacy to participate in PA’ was the most recurrent theme. Students who doubted their success in becoming physically active or who lacked confidence in their ability to initiate PA or participate in sport were less motivated to take part. A less frequent facilitator was students’ ‘self-affirmation to participate in PA’, often referring to positive cognitions about one’s own physical abilities.

11. Cognitive and interpersonal skills (Psychological Capability) ( n  = 15%)

‘Time-management’ was the only theme identified within cognitive and interpersonal skills. Students who struggled to manage their time effectively found it difficult to incorporate regular PA into their daily routine.

12. Social/professional role and identity (Reflective Motivation) ( n  = 8%)

The most frequent theme within social/professional role and identity was ‘perceiving PA as a part of one’s self-identity’. Students who engaged regularly in PA often considered it integral to their identity. Conversely, students who perceived they did not align with the aesthetic and superficial stereotypes commonly associated with the fitness industry felt less motivated to be active. A specific facilitator emerged among physiotherapy students, who were motivated to be active due to the emphasis on PA within their profession.

13. Behavioural regulation (Psychological Capability) ( n  = 3%)

Within the domain of behavioural regulation, two facilitators were equally prevalent: ‘self-monitoring of PA’ and ‘feedback on progress towards a PA-related goal’. By keeping track of their step count and receiving feedback on walking goals, students were motivated to exceed the average number of daily steps or achieve their personal PA targets.

14. Memory, attention, and decision process (Psychological Capability); Optimism (Reflective Motivation) ( n  = 0%)

No barriers or facilitators relating to the TDF domains of memory, attention and decision process, or optimism were identified.

This systematic review used the TDF and COM-B model to identify barriers and facilitators to PA among university students and rank the relative importance of each TDF domain. It is the first review to apply these frameworks in the context of increasing university students’ participation in PA. Twelve TDF domains across all six sub-components of the COM-B model were identified. The three most important TDF domains were ‘environmental context and resources’, ‘social influences’, and ‘goals’. The most common barriers and facilitators were ‘lack of time’, ‘easily accessible exercise options, facilities and equipment’, ‘exercising with others’, and ‘prioritisation of PA compared to other activities’.

The most common barrier to PA was perceived lack of time. This is consistent with previous findings among university students [ 13 , 74 ] and across other populations [ 24 ], For students, lack of time was frequently attributed to a combination of competing priorities and underdeveloped time management skills. Students predominantly prioritised study over PA, as performing well at university is a valued goal and there is a common perception that spending time exercising (at the expense of study) will impede their academic success [ 53 , 58 ]. Evidence from cognitive neuroscience research, however, suggests that this is a mistaken belief. In addition to its broad physical and mental health benefits, a growing body of evidence demonstrates regular PA can change the structure and function of the brain.

These changes can, in turn, enhance numerous aspects of cognition, including memory, attention, and processing speed [ 4 , 75 , 76 , 77 ], and buffer the negative impact of stress on cognition [ 78 ], all of which are important for academic success. However, students are typically unaware of the brain and cognitive health benefits of PA and its potential to improve academic performance, particularly compared to the physical health benefits [ 37 , 40 , 64 ]. Interventions that position participating in PA as a conduit for helping, rather than hindering, academic goals could increase the relative importance of PA to students and therefore increase their motivation to regularly engage in it. The impact that interventions of this nature have on students’ PA is yet to be empirically assessed.

Ineffective time management also contributed to students’ perceived lack of time for PA. Students reported tendencies to procrastinate in the face of overwhelming academic workloads, which left limited time for PA [ 53 ]. Additionally, students lacked an understanding of how to organise time for PA around academic timetables, social and family responsibilities, co-curricular activities, and employment commitments [ 9 , 44 , 53 , 59 ]. To address these challenges, efforts to develop students’ time management skills will be useful for enabling students to regularly participate in PA. Goal-setting and action planning are two specific examples of such skills that can be integrated into interventions to help students initiate and maintain a PA routine [ 79 ]. For example, goal-setting could involve setting a daily PA goal, and action planning could involve planning to engage in a particular PA at a particular time on certain days.

While the most common determinants of university students’ PA levels were not influenced by specific demographic characteristics, several barriers disproportionately impacted women and students with a disability. These findings are in keeping with evidence that PA is lower among these equity-deserving groups compared with the general population [ 68 , 80 ]. For women, particularly those from Middle Eastern cultures, restrictions were often tied to religious practices and sociocultural norms that limited their opportunities to engage in PA [ 45 , 48 , 66 ]. Additionally, a substantial number of women felt intimidated or self-conscious when exercising in front of others, especially men [ 48 , 49 ]. They also felt that exercise facilities were more often tailored towards the needs of men, leading to a perception that they were unwelcome in exercise communities [ 45 , 48 ]. Consequently, women expressed a desire for women-only spaces to exercise to help them overcome these gender-specific barriers to PA [ 47 , 48 , 66 ]. Furthermore, students with a disability faced physical accessibility barriers and perceived stigmatisation that deterred them from PA [ 50 , 52 ]. The lack of accessible exercise facilities and suitable equipment, programs, and education regarding how to adapt physical activities to accommodate their needs limited their opportunity and ability to participate [ 52 ]. Moreover, students with a disability felt stigmatised by others for not fitting into public perceptions of ‘normality’ or the aesthetic values and beauty standards often portrayed by the fitness industry [ 50 ]. These barriers for both equity-deserving groups of students are deeply rooted in historical stereotypes that have traditionally excluded women and people with a disability from engaging in various types of PA [ 81 , 82 ]. Despite growing awareness of these issues, PA inequalities persist due to narrow sociocultural norms, and a lack of diverse representation and inclusion in the fitness industry and associated marketing campaigns [ 83 , 84 ]. A concerted effort to address PA inequalities across the university sector and fitness industry more broadly is needed. One approach for achieving this is to develop interventions that are tailored to the unique needs of equity-deserving groups, emphasise inclusivity, diversity, and empowerment, and feature women and people with a disability being active.

The “This Girl Can” [ 85 ] and “Everyone Can” [ 86 ] multimedia campaigns are two examples of health behaviour interventions that were co-developed with key stakeholders (i.e., women and people with a disability, respectively) to tackle PA inequalities. The “This Girl Can” campaign has reached over 3 million women and girls, projecting inclusive and positive messages that aim to empower them to be physically active. Following the widespread reach of the “This Girl Can” campaign, the “Everybody Can” campaign was launched to support the inclusion of people with a disability in the PA sector. Although not tailored for university students, these campaigns provide a useful example for developing interventions that are specifically designed to address key barriers preventing women and people with a disability from participating in PA.

Across the tertiary education sector globally, efforts to elevate opportunities and motivation to include PA as a core part of the student experience will be beneficial for promoting students’ PA at scale. Two intervention approaches that can be implemented to facilitate such an endeavour are environmental restructuring and enablement [ 17 ]. These intervention approaches should involve the provision of accessible low-cost exercise options, facilities, and programs, integrating PA into the university curriculum, and mobilising student and staff leadership to encourage students’ participation in PA [ 9 ]. Although there is evidence that these approaches can be effective in promoting sustained PA throughout students’ university years and beyond [ 87 ], implementation measures such as these are complex. Implementation requires aligning student activity levels with broader university goals and is further complicated by having to compete with other funding priorities and resource allocations. Notably, due to the negative impact of the COVID-19 pandemic on university students’ physical and mental health [ 88 , 89 ], the post-pandemic era has seen many universities prioritise enhancing student health and wellbeing alongside more traditional strategic goals like academic excellence and workforce readiness. Despite the potential for PA to be used as a vehicle for supporting these strategic goals there is an absence of data on the extent to which this is occurring in the university sector. The limited evidence in this area suggests that some universities have made efforts to support students’ mental health by referring students who access on-campus counselling services to PA programs [ 90 ]. However, the uptake and efficacy of such initiatives is rarely assessed, and even less is known about whether PA is being used to support other strategic goals, such as academic success. Therefore, while the potential is there for the university sector to use PA to support students’ mental health and academic performance, to be successful this needs to become a strategic university priority. Given that these strategic priorities are set at the senior leadership level, engaging senior university staff in intervention design and promotion efforts is important to enhance the value of PA in the tertiary education sector.

Implications for intervention development

The current findings provide a high-level synthesis of the most common barriers and facilitators to university students’ physical activity. These findings can be leveraged with behavioural intervention development tools and frameworks (e.g., the BCW [ 17 ], Obesity-Related Behavioural Intervention Trials model [ 91 ], Intervention Mapping [ 92 ], and the Medical Research Council guidelines for developing complex interventions [ 93 , 94 ]) to develop evidence-based interventions and policies to promote PA. Given that the TDF and COM-B model are directly linked to the BCW framework, applying this process may be particularly useful to translate the current findings into an intervention.

Additionally, current findings can be triangulated with data directly collected from key stakeholders to assist in the development of context-specific interventions. Best practice principles for developing behavioural interventions recommend this approach to ensure a deep understanding of the barriers and facilitators that need to be targeted to increase the likelihood of behaviour change [ 17 ]. Consulting stakeholders directly (i.e., university students and staff) to understand their perspectives on the barriers and facilitators to students’ PA also enables an intervention to be appropriately tailored to the target population’s needs and implementation setting. Studies continue to demonstrate the effectiveness of this approach, especially when framed within the context of frameworks directly linked to intervention development frameworks, such as the TDF [ 95 ].

Strengths and limitations

The findings of this review should be considered with respect to its methodological strengths and limitations. The credibility and reliability of the research findings are supported by a systematic approach to screening and analysing the empirical data, along with the use of gold-standard behavioural science frameworks to classify barriers and facilitators to PA. The inclusion of qualitative, quantitative, and mixed-methods studies of both barriers and facilitators to students’ PA allowed for a comprehensive understanding of the factors that influence students’ PA that have not previously been captured.

While the present review elucidates students’ own perspectives of the factors that influence their activity levels, other stakeholders such as university staff, will also influence the adoption, operationalisation, and scale of PA interventions in a university setting. It will be important for future research to explore factors that influence university decision-makers in these roles to inform large-scale strategies for promoting students' PA.

Additionally, only one study included in the review used the TDF to explore barriers and facilitators to PA [ 47 ]. Therefore, it is possible that certain TDF domains may not have been identified because students were not asked relevant questions to assess the influence of those domains on their PA. For instance, domains such as ‘memory, attention, and decision process’, and ‘optimism’ are likely to play a role in understanding the barriers and facilitators to PA despite not being identified in this review.

Moreover, quantitative data were only extracted if ≥ 50% of students endorsed the factor as a barrier or facilitator to PA. This threshold was purposefully applied to maintain a focus on the TDF domains most universally relevant to the broad student population in the context of understanding their barriers and facilitators to PA. It is possible that less frequently reported barriers and facilitators, which may not be as prominently featured in the results, could be relevant to specific groups of students, such as those identified as equity-deserving.

Lastly, a quality appraisal of the included studies was not undertaken. This decision was informed by the aim of the review, which was to describe and synthesise the literature to subsequently map data to the TDF and COM-B rather than assess the effectiveness of interventions or determine the strength of evidence. However, this decision, combined with dual screening 25% of the studies and excluding unpublished studies and grey literature, may introduce sources of error and bias, which should be considered when interpreting the results presented.

PA is an effective, scalable, and empowering means of enhancing physical, mental, and cognitive health. This approach could help students reach their academic potential and cope with the many stressors that accompany student life, in addition to setting a strong foundation for healthy exercise habits for a lifetime. As such, understanding the barriers and facilitators to an active student lifestyle is beneficial. This systematic review applied the TDF and COM-B model to identify and map students’ barriers and facilitators to PA and, in doing so, provides a pragmatic, theory-informed, and evidence-based foundation for designing future context-specific PA interventions. The findings from this review highlight the importance of developing PA interventions that focus on the TDF domains ‘environmental context and resources’, ‘social influences’, and ‘goals’, for which intervention approaches could involve environmental restructuring, education, and enablement. If successful, such strategies could make a significant contribution to improving the overall health and academic performance of university students.

Availability of data and materials

The review protocol is available on PROSPERO. The datasets used and/or analysed during the current study and materials used are available from the corresponding author on reasonable request.

The term ‘intervention’ was included to identify student barriers and facilitators to engaging in implemented physical activity interventions.

Physical exercise is defined as “a subset of physical activity that is planned, structured, and repetitive”, and purposefully focused on the improvement or maintenance of physical fitness, whereas physical activity is defined as “any bodily movement produced by skeletal muscles that results in energy expenditure” [ 96 ].

Abbreviations

Behaviour Change Wheel

Capability, Opportunity, Model-Behaviour

  • Physical activity

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

International Prospective Register of Systematic Reviews

Theoretical Domains Framework

Naci H, Ioannidis JPA. Comparative effectiveness of exercise and drug interventions on mortality outcomes: metaepidemiological study. BMJ. 2013;347:f5577.

Article   PubMed   PubMed Central   Google Scholar  

Stensel D, Hardman A, Gill J. Physical Activity and Health: The Evidence Explained. 2021.

Book   Google Scholar  

Stubbs B, Vancampfort D, Hallgren M, Firth J, Veronese N, Solmi M, et al. EPA guidance on physical activity as a treatment for severe mental illness: a meta-review of the evidence and Position Statement from the European Psychiatric Association (EPA), supported by the International Organization of Physical Therapists in Mental Health (IOPTMH). Eur Psychiatry. 2018;54:124–44.

Article   ADS   PubMed   Google Scholar  

Hotting K, Roder B. Beneficial effects of physical exercise on neuroplasticity and cognition. Neurosci Biobehav Rev. 2013;37(9):2243–57.

Article   PubMed   Google Scholar  

World Health Organization. Global action plan on physical activity 2018–2030: more active people for a healthier world. Geneva: World Health Organization; 2018.

Keating XD, Guan J, Piñero JC, Bridges DM. A Meta-Analysis of College Students’ Physical Activity Behaviors. J Am Coll Health. 2005;54(2):116–26.

Worsley JD, Harrison P, Corcoran R. Bridging the Gap: Exploring the Unique Transition From Home, School or College Into University. Frontiers in Public Health. 2021;9. Available from: https://doi.org/10.3389/fpubh.2021.634285

Romaguera D, Tauler P, Bennasar M, Pericas J, Moreno C, Martinez S, et al. Determinants and patterns of physical activity practice among Spanish university students. J Sports Sci. 2011;29(9):989–97.

Deliens T, Deforche B, De Bourdeaudhuij I, Clarys P. Determinants of physical activity and sedentary behaviour in university students: a qualitative study using focus group discussions. BMC Public Health. 2015;15(1):201.

Buckworth J, Nigg C. Physical activity, exercise, and sedentary behavior in college students. J Am Coll Health. 2004;53:28–34.

Nelson MC, Kocos R, Lytle LA, Perry CL. Understanding the perceived determinants of weight-related behaviors in late adolescence: a qualitative analysis among college youth. J Nutr Educ Behav. 2009;41(4):287–92.

Rouse PC, Biddle SJH. An ecological momentary assessment of the physical activity and sedentary behaviour patterns of university students. Health Educ J. 2010;69(1):116–25.

Article   Google Scholar  

Ferreira Silva RM, Mendonça CR, Azevedo VD, Raoof Memon A, Noll P, Noll M. Barriers to high school and university students’ physical activity: A systematic review. Huertas-Delgado FJ, editor. PLoS ONE. 2022;17(4):e0265913.

Article   CAS   PubMed   PubMed Central   Google Scholar  

Atkins L, Francis J, Islam R, O’Connor D, Patey A, Ivers N, et al. A guide to using the theoretical domains framework of behaviour change to investigate implementation problems. Implement Sci. 2017;12(1):77.

Michie S, Johnston M, Abraham C, Lawton R, Parker D, Walker A. Making psychological theory useful for implementing evidence based practice: a consensus approach. Qual Saf Health Care. 2005;14(1):26.

Cane J, O’Connor D, Michie S. Validation of the theoretical domains framework for use in behaviour change and implementation research. Implement Sci. 2012;7(1):37.

Michie S, van Stralen MM, West R. The behaviour change wheel: a new method for characterising and designing behaviour change interventions. Implement Sci. 2011;6(1):42.

Bandura A. Social foundations of thought and action: A social cognitive theory. Social foundations of thought and action: A social cognitive theory. 1986. xiii, 617–xiii, 617.

Google Scholar  

Bandura A. Health promotion by social cognitive means. Health Educ Behav. 2004;31(2):143–64.

Ajzen I. The theory of planned behavior. Organ Behav Hum Decis Process. 1991;50(2):179–211.

Ajzen I. The theory of planned behavior. Handbook Theor Soc Psychol. 2012;1:438–59.

Atkins L, Sallis A, Chadborn T, Shaw K, Schneider A, Hopkins S, et al. Reducing catheter-associated urinary tract infections: a systematic review of barriers and facilitators and strategic behavioural analysis of interventions. Implement Sci. 2020;15:1–22.

Chung OS, Dowling NL, Brown C, Robinson T, Johnson AM, Ng CH, et al. Using the theoretical domains framework to inform the implementation of therapeutic virtual reality into mental healthcare. Administration and Policy in Mental Health and Mental Health Services Research. 2023;50(2):237–68.

Weatherson KA, McKay R, Gainforth HL, Jung ME. Barriers and facilitators to the implementation of a school-based physical activity policy in Canada: application of the theoretical domains framework. BMC Public Health. 2017;17(1):835.

Taylor N, Lawton R, Slater B, Foy R. The demonstration of a theory-based approach to the design of localized patient safety interventions. Implement Sci. 2013;8(1):123.

Garne-Dalgaard A, Mann S, Bredahl TVG, Stochkendahl MJ. Implementation strategies, and barriers and facilitators for implementation of physical activity at work: a scoping review. Chiropractic Manual Therapies. 2019;27(1):48.

Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. PLoS Med. 2021;18(3):e1003583.

Fritz T, Huang E, Murphy G, Zimmermann T. Persuasive technology in the real world: a study of long-term use of activity sensing devices for fitness. In: CHI '14 proceedings of the SIGCHI conference on human factors in computing systems: April 26–may 01, 2014; Toronto. Ontario: ACM; 2014. p. 487–96.

Tacconelli E. Systematic reviews: CRD’s guidance for undertaking reviews in health care. Lancet Infect Dis. 2010;10(4):226.

The EndNote Team. EndNote. Philadelphia, PA: Clarivate; 2013.

O’Mahony B, Kerins C, Murrin C, Kelly C. Barriers and facilitators to the implementation of nutrition standards for school food: a mixed methods systematic review protocol. HRB Open Res. 2020;3:20.

Stuart G, D’Lima D. Perceived barriers and facilitators to attendance for cervical cancer screening in EU member states: a systematic review and synthesis using the Theoretical Domains Framework. Psychol Health. 2022;37(3):279–330.

Bouma SE, van Beek JFE, Diercks RL, van der Woude LHV, Stevens M, van den Akker-Scheek I. Barriers and facilitators perceived by healthcare professionals for implementing lifestyle interventions in patients with osteoarthritis: a scoping review. BMJ Open. 2022;12(2):e056831–e056831.

Michie S, Atkins L, West R. The Behaviour Change Wheel: A Guide to Designing Interventions. London: Silverback Publishing; 2014. Available from: www.behaviourchangewheel.com .

Gale NK, Heath G, Cameron E, Rashid S, Redwood S. Using the framework method for the analysis of qualitative data in multi-disciplinary health research. BMC Med Res Methodol. 2013;13(1):117.

Braun V, Clarke V. Using thematic analysis in psychology. Qual Res Psychol. 2006;3(2):77–101.

Diehl K, Fuchs AK, Rathmann K, Hilger-Kolb J. Students’ Motivation for Sport Activity and Participation in University Sports: a Mixed-Methods Study. Biomed Res Int. 2018;2018:1–7.

Bellows-Riecken K, Mark R, Rhodes RE. Qualitative elicitation of affective beliefs related to physical activity. Psychol Sport Exerc. 2013;14(5):786–92.

Ramírez-Vélez R. Prevalencia De Barreras Para La Práctica De Actividad Física En. Nutr Hosp. 2015;2:858–65.

Lerner J, Burns C, De Róiste Á. Correlates of physical activity among college students. Recreational Sports Journal. 2011;35(2):95–106.

Snyder K, Lee JM, Bjornsen A, Dinkel D. What gets them moving? College students’ motivation for exercise: an exploratory study. Recreational Sports Journal. 2017;41(2):111–24.

Walsh A, Taylor C, Brennick D. Factors that influence campus dwelling university students’ facility to practice healthy living guidelines. Can J Nurs Res. 2018;50(2):57–63.

Forrest CK, Bruner MW. Evaluating social media as a platform for delivering a team-building exercise intervention: a pilot study. Int J Sport Exercise Psychol. 2017;15(2):190–206.

Ranasinghe C, Sigera C, Ranasinghe P, Jayawardena R, Ranasinghe ACR, Hills AP, et al. Physical inactivity among physiotherapy undergraduates: exploring the knowledge-practice gap. BMC Sports Sci Med Rehabil. 2016;8(1):39.

Othman MS, Mat Ludin AF, Chen LL, Hossain H, Abdul Halim II, Sameeha MJ, et al. Motivations, barriers and exercise preferences among female undergraduates: a need assessment analysis. Muazu Musa R, editor. PLoS ONE. 2022;17(2):e0264158.

Leinberger-Jabari A, Al-Ajlouni Y, Ieriti M, Cannie S, Mladenovic M, Ali R. Assessing motivators and barriers to active healthy living among a multicultural college student body: a qualitative inquiry. J Am Coll Health. 2023;71(2):338–42.

Burton NW, Barber BL, Khan A. A qualitative study of barriers and enablers of physical activity among female Emirati university students. IJERPH. 2021;18(7):3380.

silver mp, easty lk, sewell km, georges r, behman a. perspectives on exercise participation among canadian university students. health educ j. 2019;78(7):851–65.

lacaille lj, dauner kn, krambeer rj, pedersen j. psychosocial and environmental determinants of eating behaviors, physical activity, and weight change among college students: a qualitative analysis. J Am Coll Health. 2011;59(6):531–8.

Monforte J, Úbeda-Colomer J, Pans M, Pérez-Samaniego V, Devís-Devís J. Environmental barriers and facilitators to physical activity among university students with physical disability—a qualitative study in Spain. IJERPH. 2021;18(2):464.

Brunette MK, Lariviere M, Schinke RJ, Xing X, Pickard P. Fit to belong: activity and acculturation of Chinese students. J Sport Behav. 2011;34(3):207.

Devine MA. Leisure-time physical activity: experiences of college students with disabilities. Adapt Phys Activ Q. 2016;33(2):176–94.

Article   MathSciNet   PubMed   Google Scholar  

Kwan MYW, Faulkner GEJ. Perceptions and barriers to physical activity during the transition to university. 2011.

Pan J, Nigg C. Motivation for physical activity among hawaiian, japanese, and filipino university students in Hawaii. J Appl Sport Psychol. 2011;23(1):1–15.

Wilson OWA, Walters SR, Naylor ME, Clarke JC. University students’ negotiation of physical activity and sport participation constraints. Recreational Sports Journal. 2019;43(2):84–92.

Tong HL, Coiera E, Laranjo L. Using a mobile social networking app to promote physical activity: a qualitative study of users’ perspectives. J Med Internet Res. 2018;20(12):e11439.

Marmo J. Applying SOCIAL COGNITIVE THEORY TO DEVELOP TARGETED MESSAGES: COLLEGE STUDENTS AND PHYSICAL ACTIVITY. West J Commun. 2013;77(4):444–65.

Hilger-Kolb J, Loerbroks A, Diehl K. ‘When I have time pressure, sport is the first thing that is cancelled’: a mixed-methods study on barriers to physical activity among university students in Germany. J Sports Sci. 2020;38(21):2479–88.

Nannyonjo J, Nsibambi C, Goon D, Amusa L. Physical activity patterns of female students of Kyambogo University, Uganda. African Journal for Physical, Health Education, Recreation and Dance. 2013;19(4:1):865–73.

El Gilany AH, Badawi K, El Khawaga G, Awadalla N. Physical activity profile of students in Mansoura University. Egypt East Mediterr Health J. 2011;17(08):694–702.

Goldstein SP, Forman EM, Butryn ML, Herbert JD. Differential programming needs of college students preferring web-based versus in-person physical activity programs. Health Commun. 2018;33(12):1509–15.

Griffiths K, Moore R, Brunton J. Sport and physical activity habits, behaviours and barriers to participation in university students: an exploration by socio-economic group. Sport Educ Soc. 2022;27(3):332–46.

King KA, Vidourek RA, English L, Merianos AL. Vigorous physical activity among college students: using the health belief model to assess involvement and social support. AEHD. 2014;4(2):267–79.

Miyawaki C, Ohara K, Mase T, Kouda K, Fujitani T, Momoi K, et al. The purpose and the motivation for future practice of physical activity and related factors in Japanese university students. jhse. 2019;14(1). Available from: http://hdl.handle.net/10045/77873 .

Pellerine LP, Bray NW, Fowles JR, Furlano JA, Morava A, Nagpal TS, et al. The influence of motivators and barriers to exercise on attaining physical activity and sedentary time guidelines among Canadian undergraduate students. IJERPH. 2022;19(19):12225.

Aljayyousi GF, Abu Munshar M, Al-Salim F, Osman ER. Addressing context to understand physical activity among Muslim university students: the role of gender, family, and culture. BMC Public Health. 2019;19(1):1452.

Nolan VT, Sandada M, Surujlal J. Perceived benefits and barriers to physical exercise participation of first year university students. 2011.

Chaabna K, Mamtani R, Abraham A, Maisonneuve P, Lowenfels AB, Cheema S. Physical activity and Its barriers and facilitators among university students in Qatar: a cross-sectional study. IJERPH. 2022;19(12):7369.

Awadalla NJ, Aboelyazed AE, Hassanein MA, Khalil SN, Aftab R, Gaballa II, et al. Assessment of physical inactivity and perceived barriers to physical activity among health college students, south-western Saudi Arabia. East Mediterr Health J. 2014;20(10):596–604.

Article   CAS   PubMed   Google Scholar  

Cooke PA, Tully MA, Cupples ME, Gilliland AE, Gormley GJ. A randomised control trial of experiential learning to promote physical activity. Educ Prim Care. 2013;24(6):427–35.

Musaiger AO, Al-Kandari FI, Al-Mannai M, Al-Faraj AM, Bouriki FA, Shehab FS, et al. Perceived barriers to weight maintenance among university students in Kuwait: the role of gender and obesity. Environ Health Prev Med. 2014;19(3):207–14.

Quintiliani LM, Bishop HL, Greaney ML, Whiteley JA. Factors across home, work, and school domains influence nutrition and physical activity behaviors of nontraditional college students. Nutr Res. 2012;32(10):757–63.

Von Sommoggy J, Rueter J, Curbach J, Helten J, Tittlbach S, Loss J. How does the campus environment influence everyday physical activity? A photovoice study among students of two German universities. Front Public Health. 2020;8:561175.

Pan M, Ying B, Lai Y, Kuan G. Status and Influencing Factors of Physical Exercise among College Students in China: A Systematic Review. Int J Environ Res Public Health. 2022;19(20):13465.

Den Ouden L, Kandola A, Suo C, Hendrikse J, Costa RJS, Watt MJ, et al. The influence of aerobic exercise on hippocampal integrity and function: preliminary findings of a multi-modal imaging analysis. Brain Plast. 2018;4(2):211–6.

Maleki S, Hendrikse J, Chye Y, Caeyenberghs K, Coxon JP, Oldham S, et al. Associations of cardiorespiratory fitness and exercise with brain white matter in healthy adults: a systematic review and meta-analysis. Brain Imaging Behav. 2022;16(5):2402–25.

Müller P, Duderstadt Y, Lessmann V, Müller NG. Lactate and BDNF: key mediators of exercise induced neuroplasticity? J Clin Med. 2020;9(4):1136.

Nicastro TM, Greenwood BN. Central monoaminergic systems are a site of convergence of signals conveying the experience of exercise to brain circuits involved in cognition and emotional behavior. Curr Zool. 2016;62(3):293–306.

Fleig L, Pomp S, Parschau L, Barz M, Lange D, Schwarzer R, et al. From intentions via planning and behavior to physical exercise habits. Psychol Sport Exerc. 2013;14(5):632–9.

Carroll DD, Courtney-Long EA, Stevens AC, Sloan ML, Lullo C, Visser SN, et al. Vital signs: disability and physical activity—United States, 2009–2012. Morb Mortal Wkly Rep. 2014;63(18):407.

Ferez S. From women’s exclusion to gender institution: a brief history of the sexual categorisation process within sport. Int J History Sport. 2012;29(2):272–85.

Ferez S, Ruffié S, Joncheray H, Marcellini A, Pappous S, Richard R. Inclusion through sport: a critical view on paralympic legacy from a historical perspective. Social Inclusion. 2020;8(3):224–35.

Rasmussen K, Dufur MJ, Cope MR, Pierce H. Gender marginalization in sports participation through advertising: the case of Nike. Int J Environ Res Public Health. 2021;18(15):7759.

Richardson EV, Smith B, Papathomas A. Disability and the gym: Experiences, barriers and facilitators of gym use for individuals with physical disabilities. Disabil Rehabil. 2017;39(19):1950–7.

Sport England. This Girl Can | This Girl Can | This girl can. 2015 [cited 2023 Aug 21]. Available from: https://www.thisgirlcan.co.uk/ .

ukactive. Everyone Can. 2017 [cited 2023 Aug 21]. Available from: https://everyonecan.ukactive.com/ .

Kim M, Cardinal BJ. A Review of How Physical Activity Education Policies in Higher Education Affect College Students’ Physical Activity Behavior and Motivation. In 2016. Available from: https://api.semanticscholar.org/CorpusID:210543192 .

Ro A, Rodriguez VE, Enriquez LE. Physical and mental health impacts of the COVID-19 pandemic among college students who are undocumented or have undocumented parents. BMC Public Health. 2021;21(1):1580.

Wang X, Hegde S, Son C, Keller B, Smith A, Sasangohar F. Investigating mental health of US college students during the COVID-19 pandemic: cross-sectional survey study. J Med Internet Res. 2020;22(9):e22817.

Brunton JA, Mackintosh CI. Interpreting university sport policy in England: seeking a purpose in turbulent times? Int J Sport Policy Politics. 2017;9(3):377–95.

Czajkowski SM, Powell LH, Adler N, Naar-King S, Reynolds KD, Hunter CM, et al. From ideas to efficacy: the ORBIT model for developing behavioral treatments for chronic diseases. Health Psychol. 2015;34(10):971.

Eldredge LKB, Markham CM, Ruiter RA, Fernández ME, Kok G, Parcel GS. Planning health promotion programmes: an intervention mapping approach. 4th ed. San Francisco (USA): Wiley; 2016.

Craig P, Dieppe P, Macintyre S, Michie S, Nazareth I, Petticrew M. Developing and evaluating complex interventions: the new Medical Research Council guidance. BMJ. 2008;337(7676):979–83.

Skivington K, Matthews L, Simpson SA, Craig P, Baird J, Blazeby JM, et al. A new framework for developing and evaluating complex interventions: update of Medical Research Council guidance. BMJ. 2021;374:n2061.

Ndupu LB, Staples V, Lipka S, Faghy M, Bessadet N, Bussell C. Application of theoretical domains framework to explore the enablers and barriers to physical activity among university staff and students: a qualitative study. BMC Public Health. 2023;23(1):670–670.

Caspersen CJ, Powell KE, Christenson GM. Physical activity, exercise, and physical fitness: definitions and distinctions for health-related research. Public Health Rep. 1985;100(2):126–31.

CAS   PubMed   PubMed Central   Google Scholar  

Download references

Acknowledgements

The authors extend their gratitude to the funder, the nib foundation, for its financial support, which was instrumental in facilitating this research. We are also indebted to the Wilson Foundation and the David Winston Turner Endowment Fund for their generous philanthropic contributions, which have supported the BrainPark research team and facility where this research was conducted. Special thanks are owed to the library staff at Monash University for their expertise in conducting systematic reviews, which helped inform the selection of databases and the development of the search strategy.

This research was supported by nib foundation. The nib foundation had no role in the design of the study and collection, analysis, and interpretation of data, and in writing the manuscript. The views expressed are those of the authors and not necessarily those of the nib foundation.

Author information

Murat Yücel and Rebecca A. Segrave share senior authorship.

Authors and Affiliations

BrainPark, Turner Institute for Brain and Mental Health, Monash University, Melbourne, Australia

Catherine E. B. Brown, Karyn Richardson, Bengianni Halil-Pizzirani & Rebecca A. Segrave

Centre for Behaviour Change, University College London, London, UK

QIMR Berghofer Medical Research Institute, Herston, Brisbane, QLD, Australia

Murat Yücel

You can also search for this author in PubMed   Google Scholar

Contributions

CB, KR, BP, LA and RS developed the review protocol. CB and BP conducted the search and screened articles, and KR resolved conflicts. CB, KR, BP, LA and RS extracted the barriers and facilitators, mapped barriers and facilitators to the TDF and COM-B model, and interpreted the results. CB drafted the paper. All authors read, revised, and approved the submitted version.

Corresponding author

Correspondence to Catherine E. B. Brown .

Ethics declarations

Ethics approval and consent to participate.

The current review contained no participants and did not require ethical approval.

Consent for publication

No individual person’s data in any form is contained in the current article.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1. .

Theoretical Domains Framework domains, definitions, and constructs.

Additional file 2. 

Search syntax for Ovid MEDLINE.

Additional file 3. 

Summary of study characteristics.

Additional file 4. 

PRISMA Checklist.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Brown, C.E.B., Richardson, K., Halil-Pizzirani, B. et al. Key influences on university students’ physical activity: a systematic review using the Theoretical Domains Framework and the COM-B model of human behaviour. BMC Public Health 24 , 418 (2024). https://doi.org/10.1186/s12889-023-17621-4

Download citation

Received : 26 August 2023

Accepted : 30 December 2023

Published : 09 February 2024

DOI : https://doi.org/10.1186/s12889-023-17621-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

  • University students
  • Physical exercise
  • Behaviour change
  • Theoretical domains framework
  • Facilitators

BMC Public Health

ISSN: 1471-2458

analysis frame research

analysis frame research

Dalton Transactions

Uranyl uptake into metal–organic frameworks: a detailed x-ray structural analysis.

Metal-organic frameworks (MOF) are a subclass of porous framework materials that have been used for a wide variety of applications in sensing, catalysis, and remediation. Among these myriad applications is their remarkable ability to capture substances in a variety of environments ranging from benign to extreme. Among the most common and problematic substances found throughout the world’s oceans and water supplies is [UO 2 ] 2+ , a common mobile ion of uranium, which is found both naturally and as a result of anthropogenic activities, leading to problematic environmental contamination. While some MOFs possess high capability for the uptake of [UO 2 ] 2+ , many more of the thousands of MOFs and their modifications that have been produced over the years have yet to be studied for their ability to uptake [UO 2 ] 2+ . However, studying the thousands of MOFs and their modifications presents an incredibly difficult task. As such, a way to narrow down the numbers seems imperative. Herein, we evaluate the binding behaviors as well as identify the specific binding sites of [UO 2 ] 2+ incorporated into six different Zr MOFs to elucidate specific features that improve [UO 2 ] 2+ uptake. In doing so, we also present a method for the determination and verification of these binding sites by Anomalous wide-angle X-ray scattering, X-ray fluorescence, and X-ray absorption spectroscopy. This research not only presents a way for future research into the uptake of [UO 2 ] 2+ into MOFs to be conducted but also a means to evaluate MOFs more generally for the uptake of other compounds to be applied for environmental remediation and improvement of ecosystems globally.

Supplementary files

  • Supplementary information PDF (2748K)

Article information

Download citation, permissions.

analysis frame research

M. P. Heaney, H. M. Johnson, J. G. Knapp, S. Bang, S. Seifert, N. S. Yaw, J. Li, O. K. Farha, Q. Zhang and L. M. Moreau, Dalton Trans. , 2024, Accepted Manuscript , DOI: 10.1039/D3DT04284G

To request permission to reproduce material from this article, please go to the Copyright Clearance Center request page .

If you are an author contributing to an RSC publication, you do not need to request permission provided correct acknowledgement is given.

If you are the author of this article, you do not need to request permission to reproduce figures and diagrams provided correct acknowledgement is given. If you want to reproduce the whole article in a third-party publication (excluding your thesis/dissertation for which permission is not required) please go to the Copyright Clearance Center request page .

Read more about how to correctly acknowledge RSC content .

Social activity

Search articles by author.

This article has not yet been cited.

Advertisements

  • Search Menu
  • Supplements
  • Advance articles
  • Editor's Choice
  • Special Issues
  • Author Guidelines
  • Submission Site
  • Why Publish With Us?
  • Open Access
  • About Nicotine & Tobacco Research
  • About Society for Nicotine & Tobacco Research
  • Editorial Board
  • Advertising and Corporate Services
  • Journals Career Network
  • Self-Archiving Policy
  • Dispatch Dates
  • Journals on Oxford Academic
  • Books on Oxford Academic

Article Contents

  • Meta-Analysis
  • Declaration of Interests
  • Author Contributions

The Impact of Menthol Cigarette Bans: A Systematic Review and Meta-Analysis

ORCID logo

  • Article contents
  • Figures & tables
  • Supplementary Data

Sarah D Mills, Snigdha Peddireddy, Rachel Kurtzman, Frantasia Hill, Victor Catalan, Jennifer S Bissram, Kurt M Ribisl, The Impact of Menthol Cigarette Bans: A Systematic Review and Meta-Analysis, Nicotine & Tobacco Research , 2024;, ntae011, https://doi.org/10.1093/ntr/ntae011

  • Permissions Icon Permissions

This review investigates the impacts of banning the sale of menthol cigarettes at stores.

A systematic search of studies published in English up to November 2022 was conducted. The following databases were searched: PubMed/Medline, CINAHL, PsycINFO, Web of Science, and Embase, as well as a non-indexed journal. Studies evaluating either the impact of real-world or hypothesized menthol cigarette bans were included. Primary outcomes include tobacco use behaviors. Secondary outcomes include cigarette sales, retailer compliance, and the tobacco industry’s response to a menthol ban. Data on tobacco use behavior after a menthol ban were pooled using random-effects models. Two pairs of reviewers independently extracted data and assessed study quality.

Of the 964 articles that were identified during the initial search, 78 were included in the review and 16 were included in the meta-analysis. Cessation rates among menthol cigarette smokers were high after a menthol ban. Pooled results show that 24% (95% confidence interval [95% CI]: 20%, 28%) of menthol cigarette smokers quit smoking after a menthol ban, 50% (95% CI: 31%, 68%) switched to non-menthol cigarettes, 12% (95% CI: 3%, 20%) switched to other flavored tobacco products, and 24% (95% CI: 17%, 31%) continued smoking menthol cigarettes. Hypothesized quitting and switching rates were fairly close to real-world rates. Studies found the tobacco industry attempts to undermine menthol bans. National menthol bans appear more effective than local or state menthol bans.

Menthol cigarette bans promote smoking cessation suggesting their potential to improve public health.

Findings from this review suggest that menthol cigarette bans promote smoking cessation among menthol cigarette smokers and have the potential to improve public health.

  • tobacco use
  • menthol cigarettes
  • flavored tobacco products

Email alerts

Citing articles via.

  • About Nicotine & Tobacco Research
  • Recommend to your Library

Affiliations

  • Online ISSN 1469-994X
  • Copyright © 2024 Society for Research on Nicotine and Tobacco
  • About Oxford Academic
  • Publish journals with us
  • University press partners
  • What we publish
  • New features  
  • Open access
  • Institutional account management
  • Rights and permissions
  • Get help with access
  • Accessibility
  • Advertising
  • Media enquiries
  • Oxford University Press
  • Oxford Languages
  • University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

  • Copyright © 2024 Oxford University Press
  • Cookie settings
  • Cookie policy
  • Privacy policy
  • Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

Help | Advanced Search

Computer Science > Computer Vision and Pattern Recognition

Title: benchcloudvision: a benchmark analysis of deep learning approaches for cloud detection and segmentation in remote sensing imagery.

Abstract: Satellites equipped with optical sensors capture high-resolution imagery, providing valuable insights into various environmental phenomena. In recent years, there has been a surge of research focused on addressing some challenges in remote sensing, ranging from water detection in diverse landscapes to the segmentation of mountainous and terrains. Ongoing investigations goals to enhance the precision and efficiency of satellite imagery analysis. Especially, there is a growing emphasis on developing methodologies for accurate water body detection, snow and clouds, important for environmental monitoring, resource management, and disaster response. Within this context, this paper focus on the cloud segmentation from remote sensing imagery. Accurate remote sensing data analysis can be challenging due to the presence of clouds in optical sensor-based applications. The quality of resulting products such as applications and research is directly impacted by cloud detection, which plays a key role in the remote sensing data processing pipeline. This paper examines seven cutting-edge semantic segmentation and detection algorithms applied to clouds identification, conducting a benchmark analysis to evaluate their architectural approaches and identify the most performing ones. To increase the model's adaptability, critical elements including the type of imagery and the amount of spectral bands used during training are analyzed. Additionally, this research tries to produce machine learning algorithms that can perform cloud segmentation using only a few spectral bands, including RGB and RGBN-IR combinations. The model's flexibility for a variety of applications and user scenarios is assessed by using imagery from Sentinel-2 and Landsat-8 as datasets. This benchmark can be reproduced using the material from this github link: \url{ this https URL \_segmentation\_comparative}.

Submission history

Access paper:.

  • Download PDF
  • HTML (experimental)
  • Other Formats

References & Citations

  • Google Scholar
  • Semantic Scholar

BibTeX formatted citation

BibSonomy logo

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

  • Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

COMMENTS

  1. Frame analysis

    Frame analysis (also called framing analysis) is a multi-disciplinary social science research method used to analyze how people understand situations and activities. Frame analysis looks at images, stereotypes, metaphors, actors, messages, and more. It examines how important these factors are and how and why they are chosen. [1]

  2. Analyzing frame analysis: A critical review of framing studies in

    This critical review of more than three decades of studies of frames and framing in Social Movement (SM) research first offers a brief history of the notion of 'frame' in various disciplines, and then discusses empirical studies of frame alignment, frame disputes, frame resonance and master frames, among other notions.

  3. PDF Methods of Analysis Frame Analysis

    Stephen Petrina (2016) Bateson (1954) discerned that a frame— expressed in a message, sign (signal), or statement— has implicit rules governing relations to events or objects. Language "bears to the objects which it denotes," Bateson says, "a relationship comparable to that which a map bears to a territory" (p. 69).

  4. Making the most of the frame: developing the analytical potential of

    We suggest two analytical dimensions that help to develop the analytical potential of frame analysis in policy analysis and beyond: firstly, the institutionalization process of frames which can capture the struggle of meaning within policy processes and also distinguish between the varying influences of different frames over space and time.

  5. Chapter 5: Five main stages in framework analysis

    Framework analysis describes this stage as a process of rearranging the data and thematic framework to create order, not dissimilar to the iterative principle of grounded theory. 449 We applied this stage as a principle for synthesising and developing our final coding framework through a process of abstraction, in order to derive all the detail ...

  6. Putting Framing in Perspective: A Review of Framing and Frame Analysis

    AbstractThere are few constructs that are as ubiquitous across traditions of management and organizational research, and indeed the social sciences more generally, as that of frame or framing. The widespread popularity and use of the concept has led to rich streams of research, ranging from work in managerial cognition and decision-making to strategic and organizational change, and social ...

  7. Frame Analysis

    Frame analysis is a method that describes how frames are produced, and mobilized, as well as how they interact with other frames. It reveals how individuals position themselves in relation to existing frames and identifies frames's consequences in specific contexts.

  8. Frame analysis

    Scholars in many disciplines of the humanities and social sciences have used the notion of 'frame' to analyze language, discourse, interaction, cognition, news, interaction, and social movements, thus showing the multidisciplinary conceptual relevance of the notion.

  9. Frame Analysis

    The concept of "frame" is a useful tool to explicate how people interpret and at the same time construct their interactions coherently in specific interactional situations. The concept of "frame" is traced back to Bateson, and it gets its theoretical complexity in Goffman's elaboration of "frame" and his notions of "keying" and ...

  10. Frame analysis

    Scholars in many disciplines of the humanities and social sciences have used the notion of 'frame' to analyze language, discourse, interaction, cognition, news, interaction, and social movements, thus showing the multidisciplinary conceptual relevance of the notion.

  11. Framework Analysis

    The analytical framework is a set of predetermined themes or categories that are derived from the research questions or objectives. The framework provides a structured approach to data analysis and can help to identify patterns, themes, and relationships in the data. Steps in Framework Analysis Here are the general steps in Framework Analysis:

  12. Frame analysis

    Goffman envisioned frame analysis to be an element of ethnographic research that would allow analysts to read identifiable chunks of social behaviour, or "strips," in order to understand the frames that participants use to make sense of the behaviour (whether they apprehend their reality, for instance, through a religious or a secular frame).

  13. Analyzing Talk and Text IV: Frame Analysis

    Categories for frame identification, 3. Text analysis, 4. Frame (re-)construction and contextualization. These steps are presented here in an order resulting from the requirement to explain frame analysis systematically. In qualitative frame analysis, the research process is more flexible than in quantitative research.

  14. Discourse and Frame Analysis: In-Depth Analysis of Qualitative Data in

    Introduction to Discourse and Frame Analysis in Social Movement Research The linguistic or cognitive turn that swept the humanities and the social sciences in the 1970s hit the subfield of social movement studies in the early 1980s as an evolving criticism of the established paradigm of resource mobilization theory ( McCarthy and Zald 1977 ).

  15. Analyzing frame analysis: A critical review of framing studies in

    This critical review of more than three decades of studies of frames and framing in Social Movement (SM) research first offers a brief history of the notion of 'frame' in various disciplines, and then discusses empirical studies of frame alignment, frame disputes, frame resonance and master frames, among other notions. It is found that the very notion of discursive of cognitive frames ...

  16. Making the most of the frame: developing the analytical potential of

    Two key tasks of frame analysis points of critique and a way forward. Frame analysis often involves two key analytical tasks uncovering the construction of. -. meaning within a given context, or the framing process, and identifying the effects frames have on, for example, actorsbehavior and political outcomes. Snow ' and Benford.

  17. 4722 PDFs

    Explore the latest full-text research PDFs, articles, conference papers, preprints and more on FRAME ANALYSIS. Find methods information, sources, references or conduct a literature review on FRAME ...

  18. Understanding Framework Analysis: An Introductory Guide

    Framework analysis is a specific qualitative technique initially developed by Liz Spencer, Jane Ritchie, Jane Lewis, and Lucy Dillon for the National Centre for Social Research to help the UK government with evidence-based policy making method. These techniques are used for government social research to ensure quality standards in qualitative ...

  19. Frame Analysis

    Frame analysis is the second and last analysis component in person-oriented framing analysis (PFA). The input to frame analysis is a set of news articles reporting on the same policy event, including the persons involved in the event and their mentions across all articles. The output of the frame analysis component should be information ...

  20. (PDF) Using the framework approach to analyse qualitative data: a

    ... Framework analysis [37] [38] [39] was used to organize the codes into a matrix showing the intersection between the node of interest, the level associated with that node (Table 1), and the...

  21. (PDF) Frame Analysis

    Abstract and Figures. This chapter details the last component of person-oriented framing analysis: frame analysis. The component aims to classify how persons are portrayed in news articles. The ...

  22. Framing the FRAM: A literature review on the functional resonance

    The FRAM's strengths and limitations and potential future research are presented. ... 2008), where the FRAM up to step 3 constitutes part of the framework itself. Such analysis results in a state space representation as a graphic method for representing the change in state of process variables over time, visualizing behaviour in command and ...

  23. Experiences of delivering and receiving mental healthcare in the acute

    The semi-structured interviews were interpreted independently by DR and EG using framework analysis . This is a qualitative method that is useful in research that has specific questions, a limited time frame and a pre-determined sample; it is therefore well-suited to applied policy research.

  24. Key influences on university students' physical activity: a systematic

    Data regarding study design and key findings (i.e., participant quotes, qualitative theme descriptions, and survey results) were extracted. Framework analysis was used to code barriers/facilitators to the TDF and COM-B model. Within each TDF domain, thematic analysis was used to group similar barriers/facilitators into descriptive theme labels.

  25. Uranyl Uptake into Metal-Organic Frameworks: A Detailed X-Ray

    Metal-organic frameworks (MOF) are a subclass of porous framework materials that have been used for a wide variety of applications in sensing, catalysis, and remediation. ... This research not only presents a way for future research into the uptake of [UO 2] ... A Detailed X-Ray Structural Analysis M. P. Heaney, H. M. Johnson, J. G. Knapp, S ...

  26. Impact of Menthol Cigarette Bans: A Systematic Review and Meta-Analysis

    To the best of our knowledge, this study is the first systematic review to incorporate a meta-analysis of research examining the impact of menthol cigarette bans (also referred to as menthol bans) on tobacco use behaviors. Secondary outcomes for the review include impacts on cigarette sales, retailer compliance, and the tobacco industry's ...

  27. [2402.05929] An Interactive Agent Foundation Model

    The development of artificial intelligence systems is transitioning from creating static, task-specific models to dynamic, agent-based systems capable of performing well in a wide range of applications. We propose an Interactive Agent Foundation Model that uses a novel multi-task agent training paradigm for training AI agents across a wide range of domains, datasets, and tasks. Our training ...

  28. Short-time aerodynamic research for large scale test models in

    Current research in hypersonic wind tunnel aerodynamic testing predominantly focuses on small-scale and lightweight models, typically limited to scales of 4 m and weights under 1 ton. However, there's an emerging trend in the design of hypersonic vehicles towards larger scales and heavier loads, often surpassing 9 m in length and several tons in weight. This shift poses new challenges as there ...

  29. [2402.13918] BenchCloudVision: A Benchmark Analysis of Deep Learning

    Satellites equipped with optical sensors capture high-resolution imagery, providing valuable insights into various environmental phenomena. In recent years, there has been a surge of research focused on addressing some challenges in remote sensing, ranging from water detection in diverse landscapes to the segmentation of mountainous and terrains. Ongoing investigations goals to enhance the ...

  30. Learning to Do Qualitative Data Analysis: A Starting Point

    For many researchers unfamiliar with qualitative research, determining how to conduct qualitative analyses is often quite challenging. Part of this challenge is due to the seemingly limitless approaches that a qualitative researcher might leverage, as well as simply learning to think like a qualitative researcher when analyzing data. From framework analysis (Ritchie & Spencer, 1994) to content ...