Veracity Assessment¶
Veracity assessment refers to the evaluation of the credibility, truthfulness, and factuality of news articles, claims, and sources. This encompasses both source-level assessment (evaluating the reliability of news outlets and publishers) and claim-level or article-level assessment (determining whether specific statements or articles are true, false, or unverifiable).
Veracity assessment is central to fake news research and misinformation mitigation. It forms the basis for ground truth labels in datasets used to train detection systems, and it informs public understanding of media credibility.
Key approaches¶
Source-level credibility assessment: Evaluating news outlets based on their historical accuracy, editorial standards, and track record. Organizations like Media Bias/Fact Check, NewsGuard, and AllSides provide source-level labels used in many datasets.
Article-level fact-checking: Professional fact-checkers (PolitiFact, Snopes) verify individual claims and articles, providing fine-grained ground truth.
Automated credibility scoring: Machine-learning approaches that predict veracity based on linguistic, network, and propagation features.
Key papers and resources¶
- Zubiaga et al. (2018) — Detection and Resolution of Rumours in Social Media: A Survey — comprehensive survey of veracity classification in rumour verification, reviewing datasets, annotation schemes, and approaches from feature engineering to deep learning
- Kochkina et al. (2018) — All-in-one: Multi-task Learning for Rumour Verification — demonstrates multi-task learning approach for improving veracity classification on Twitter rumour threads via joint training with stance and detection
- Gruppi, Horne & Adalı (2022) — NELA-GT-2022 — Large-scale dataset with source-level Media Bias/Fact Check labels (0–5 factuality score and conspiracy/pseudoscience classification) enabling comparative veracity assessment across 361 outlets
- Wang (2017) — Liar, Liar Pants on Fire — 6-way fine-grained statement-level labels from PolitiFact
- Thorne et al. (2018) — FEVER — Fact verification via evidence retrieval and natural language inference