User comments for fake news detection

User comments posted on news articles or social media represent an untapped signal for detecting misinformation. Comments capture:

• Skepticism: Readers questioning the veracity of claims; fact-checking assertions; pointing out logical flaws.
• Engagement patterns: Fake news may attract emotionally-charged or partisan responses, differing from real-news comments.
• Community knowledge: Aggregated reader expertise and domain familiarity, e.g., scientists debunking health misinformation.
• Temporal dynamics: Comments evolve over time as more readers encounter the article and react.

Methods exploiting comments typically encode them as learned representations (embeddings or RNNs) and combine with news content features via attention mechanisms or concatenation. Comments are valuable when available (immediately-shared articles on social platforms, news websites with active readers) but unavailable at publication time or in breaking-news scenarios.

Key papers

• Shu et al. (2019) — dEFEND: Jointly encodes news sentences and user comments via bidirectional RNNs; sentence-comment co-attention models mutual influence; outperforms content-only and comment-only baselines, suggesting complementarity.

Connections

• Social-context detection more broadly uses user engagement and network structure; comments are one dimension of social signals.
• User profiles capture demographic and behavioral features of comment authors; combining author credibility with comment content is a natural extension.
• Explainable detection uses important comments to surface reasons why a piece of news is flagged as fake.