Evidence-based Fake News Detection

Methods that assess the credibility of claims by retrieving and analyzing supporting or refuting evidence from external sources such as web articles, knowledge bases, Wikipedia, or fact-checking websites. This approach contrasts with text-only detection that relies solely on intrinsic linguistic features of the claim itself.

Key papers

• Quelle & Bovet (2023) — The Perils & Promises of Fact-checking with Large Language Models — Demonstrates that providing evidence retrieved via Google Search to GPT-3.5 and GPT-4 substantially improves fact-checking accuracy by 10–20 percentage points; uses ReAct agent framework enabling iterative search and reasoning; shows evidence integration improves both overall accuracy and calibration on ambiguous verdict categories; highlights importance of evidence quantity and quality in automated verification
• Augenstein et al. (2019) — MultiFC: A Real-World Multi-Domain Dataset for Evidence-Based Fact Checking of Claims — 34,918 naturally occurring claims from 26 domains with retrieved evidence pages and entity linking; demonstrates that jointly ranking evidence pages improves veracity prediction; best model achieves 49.2% Macro F1
• DeClareE: Debunking Fake News and False Claims using Evidence-Aware Deep Learning — end-to-end neural network combining claims and evidence articles via bidirectional LSTMs with claim-specific attention for credibility assessment.
• FEVER: A Large-Scale Dataset for Fact Extraction and VERification — benchmark dataset for fact extraction and verification pairing claims with Wikipedia evidence.
• The Fact Extraction and VERification (FEVER) Shared Task — shared task combining evidence retrieval and natural language inference for structured fact verification.
• Vo & Lee (2021) — Hierarchical Multi-head Attentive Network for Evidence-aware Fake News Detection: Jointly applies hierarchical multi-head attention at word and document levels to fact-check claims against external evidence. Word attention captures semantic importance of evidence phrases; document attention weights evidence sources by relevance. Achieves 9.47% improvement on Snopes dataset with comprehensive ablation studies.
• Jin et al. (2021) — Towards Fine-Grained Reasoning for Fake News Detection: Constructs claim-evidence graphs from social media context (posts, users, keywords) and uses mutual-reinforcement-based ranking to identify salient evidence, then reasons over evidence groups via graph neural networks. Integrates human knowledge about attribute salience to rank evidence by importance; achieves 91.7% F1 on PolitiFact while providing interpretable explanations of which evidence matters most.

Related topics

• Credibility assessment for fake news detection (broader category)
• Knowledge graphs (structured evidence sources)
• Fact-checking and corrections (manual evidence-based verification)