Political bias in fake news detection

Political bias refers to the ideological leaning of a news source, user, or content, scored on a continuous scale (typically left–right). In the context of fake news research, political bias appears as both a confound and a signal: fake news prevalence varies across partisan contexts, and users' political alignment correlates with their news-sharing behavior.

In user-profile analysis, political bias is typically inferred by computing a user's interest similarity to accounts known to be partisan (e.g., political party accounts, politically-coded hashtag communities). Scores range from strongly right-leaning (\(-1\)) to strongly left-leaning (\(+1\)).

Key empirical observations (on US political / entertainment datasets): - Users more likely to share fake news skew right-leaning; real-news sharers skew toward ideological neutrality. - Political bias is the third most important feature (Gini 0.063) after account age and verified status in user-profile classifiers on FakeNewsNet.

Caution on generalization: These findings are specific to datasets centered on US politics (PolitiFact) and entertainment (GossipCop). Political bias as a fake-news signal may not transfer to other national or topical contexts.

Key papers

• Measuring Political Bias in Large Language Models: What Is Said and How It Is Said: framework for measuring political bias in LLM-generated content, decomposing bias into stance and framing components
• A Survey on Predicting the Factuality and the Bias of News Media: comprehensive survey of media-level bias prediction, covering framing, news slant, and approaches using textual, multimedia, audience, and infrastructure features
• Allcott & Gentzkov (2017) — Social Media and Fake News in the 2016 Election: documents partisan asymmetry in fake news production and shows that political ideology predicts both belief in and sharing of fake news
• Shu et al. (2019) — The Role of User Profiles for Fake News Detection: measures political bias via interest-similarity scoring; finds it the third-ranked UPF feature by Gini importance.

Connections

• User profiles: political bias is an implicit user-profile feature.
• Social-context detection: political bias is one signal within the broader social-context paradigm.