Summary
Health discussions in social media videos often contain many interwoven statements about different health topics, opinions, and types of evidence within short conversations. The authors found that on average, people make about 13 clear health claims per minute in these videos, which shows the need for closer examination of each claim. They created a detailed system that links each claim to its topic, opinion, and other speech details, and tested it on 60 videos across four health areas. They showed that current language models can identify topics and opinions well but have trouble understanding more complex communication features. The work suggests future tools may need to break down tasks and use specialized reasoning for better results.
What this means in practice
- •For health data analysts: Enhance extraction of detailed health claims and opinions from social media videos for better health communication analysis.
- •For content moderation teams: Identify nuanced misinformation or conflicting health claims in video discussions by analyzing claim-level structures and stances.
Authors
Farnoushsadat Nilizadeh, Elham Pourabbas Vafa, Shirin Nilizadeh, Eduard Dragut
Abstract
Health discourse in social media videos often contains densely entangled claims spanning multiple thematic aspects, stances, evidential frames, and rhetorical functions within short conversational spans. Existing approaches largely rely on coarse topic-level, sentiment-based, or stance-oriented representations that do not adequately capture this structure. Our analysis identifies an average of 13.22 atomic claims per minute, motivating richer claim-level discourse representations. We introduce a structured framework for claim-level discourse analysis in dense health narratives. Our framework models discourse through tuples linking atomic claims with thematic aspects, stance, and multidimensional pragmatic discourse attributes. To support this setting, we construct a benchmark spanning four health domains with 1,191 manually annotated claims from 60 videos. Using this framework, we evaluate automated structured discourse analysis under different discourse context settings. Results show that current LLMs achieve strong performance on thematic categorization and stance prediction, but struggle with high-dimensional pragmatic profiling. We also find that different discourse tasks benefit from different forms of contextual reasoning, suggesting that future systems may require task decomposition and specialized inference strategies.