Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Framing Analysis of Health-Related Narratives: Conspiracy versus Mainstream Media

Markus Reiter-Haas, Beate Klösch, Markus Hadler, Elisabeth Lex

TL;DR

The paper tackles the challenge of analyzing health-related narrative framing by moving beyond word-frequency methods to an AMR-based approach that extracts narrative elements (characters, plot, setting, moral) from semantic graphs. Using the LOCO dataset, it contrasts conspiracy and mainstream media across COVID-19, diseases, and pharmacology, revealing that conspiracy content tends to argue from beliefs while mainstream content leans on science. The authors present a three-component pipeline (AMR parsing, narrative-element mining, and narrative-analysis) and demonstrate how smoothed log-odds contrasts and UMAP visualizations reveal systematic differences in how frames are constructed and deployed. This narrative-centric framing analysis provides a interpretable, extensible tool for understanding competing frames in online health discourse and informs efforts to counter misinformation.

Abstract

Understanding how online media frame issues is crucial due to their impact on public opinion. Research on framing using natural language processing techniques mainly focuses on specific content features in messages and neglects their narrative elements. Also, the distinction between framing in different sources remains an understudied problem. We address those issues and investigate how the framing of health-related topics, such as COVID-19 and other diseases, differs between conspiracy and mainstream websites. We incorporate narrative information into the framing analysis by introducing a novel frame extraction approach based on semantic graphs. We find that health-related narratives in conspiracy media are predominantly framed in terms of beliefs, while mainstream media tend to present them in terms of science. We hope our work offers new ways for a more nuanced frame analysis.

Framing Analysis of Health-Related Narratives: Conspiracy versus Mainstream Media

TL;DR

The paper tackles the challenge of analyzing health-related narrative framing by moving beyond word-frequency methods to an AMR-based approach that extracts narrative elements (characters, plot, setting, moral) from semantic graphs. Using the LOCO dataset, it contrasts conspiracy and mainstream media across COVID-19, diseases, and pharmacology, revealing that conspiracy content tends to argue from beliefs while mainstream content leans on science. The authors present a three-component pipeline (AMR parsing, narrative-element mining, and narrative-analysis) and demonstrate how smoothed log-odds contrasts and UMAP visualizations reveal systematic differences in how frames are constructed and deployed. This narrative-centric framing analysis provides a interpretable, extensible tool for understanding competing frames in online health discourse and informs efforts to counter misinformation.

Abstract

Understanding how online media frame issues is crucial due to their impact on public opinion. Research on framing using natural language processing techniques mainly focuses on specific content features in messages and neglects their narrative elements. Also, the distinction between framing in different sources remains an understudied problem. We address those issues and investigate how the framing of health-related topics, such as COVID-19 and other diseases, differs between conspiracy and mainstream websites. We incorporate narrative information into the framing analysis by introducing a novel frame extraction approach based on semantic graphs. We find that health-related narratives in conspiracy media are predominantly framed in terms of beliefs, while mainstream media tend to present them in terms of science. We hope our work offers new ways for a more nuanced frame analysis.
Paper Structure (27 sections, 1 equation, 4 figures, 2 tables)

This paper contains 27 sections, 1 equation, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Narration.
  4. Computational frame extraction.
  5. Frame Semantics.
  6. Notation and Definitions
  7. Method
  8. AMR Parsing
  9. Mining Narrative Elements
  10. Characters.
  11. Plot.
  12. Setting.
  13. Moral of the Story.
  14. Analysis of Narrative Information
  15. Visualization for elements.
  16. ...and 12 more sections

Figures (4)

  • Figure 1: Example sentence (top) with its extracted AMR graph using a BART-based model. Given this representation, we can identify the narrative elements, while syntactical information such as tenses is omitted. Within the narrative, three characters are present, i.e., a doctor who acts twice (i.e., two ARG0 relations) as character (orange), as well as a company and a virus (both with ARG1 relations). The plot (blue; predicates with word senses) revolves around three frames, namely prevent, vaccinate, and spread. Additionally, the year 2021 (i.e., date-entity = setting; green) and the company name Pfizer are depicted as entities (purple).
  • Figure 2: Details of the LOCO dataset in terms of temporality and distribution.
  • Figure 3: Over-represented narrative elements (i.e., plot, characters, setting, moral of the story) on COVID-19 in conspiracy versus mainstream media. Positioning is according to 2-dimensional UMAP embedding of the AMR input layer (i.e., semantically similar words appear in similar locations), and labels are force-adjusted for readability (with lines indicating their associated positioning if moved beyond a threshold).
  • Figure 4: AMR-based Framing Analysis Approach Overview