FakeNewsNet

FakeNewsNet is a benchmark data repository for fake news research, combining news content with rich social context. News labels are annotated by professional fact-checkers from two platforms:

• PolitiFact: US political news; 361 fake + 361 real news items with ~159k users and ~271k sharing events (before bot filtering).
• GossipCop: Entertainment/celebrity news; 4,513 fake + 4,513 real news items with ~210k users and ~812k sharing events.

Each news item includes article text and social context: user posting and sharing events on Twitter, enabling both content-based and social-context-based detection experiments. The dataset is dynamic — social context accumulates over time.

Papers in this wiki that use FakeNewsNet

• Shu et al. (2019) — The Role of User Profiles for Fake News Detection
• Zhou et al. (2020) — SAFE: Similarity-Aware Multi-Modal Fake News Detection
• Zhou et al. (2020) — Fake News Early Detection: An Interdisciplinary Study (old-version: PolitiFact 240 articles, BuzzFeed 180 articles)
• A Heuristic-driven Uncertainty based Ensemble Framework for Fake News Detection in Tweets and News Articles — Ensemble of pre-trained language models evaluated on FakeNewsNet dataset; achieves F1=0.9156 by combining soft voting, Statistical Feature Fusion Network, and heuristic post-processing

Notes

Bot accounts constitute 14–21% of users depending on partition; Botometer is commonly applied to filter them before profile-based analysis. The dataset has a class-balanced design (equal fake and real news counts per platform), which simplifies evaluation but may not reflect real-world distributions.