Analyzing LLM's fact-checking behaviors, testing when Internal priors and external evidence disagree. (TWINE)
Abstract
The crawled Wikipedia events (2024–2025) and a curated politifact_conflict.jsonl dataset. The complete datasets will be made publicly available upon acceptance. Wikipedia Recent Events (2024–2025): This dataset contains automatically crawled events featuring the following primary fields: Facts provides a concise summary of the event; Evidence offers detailed supporting context; and Time-year/month/day specifies the precise date. Additionally, Category labels the event type (e.g., Armed conflicts and attacks), and Resources links to the original source. Politifact Conflict Dataset: Each entry contains a claim_text, its cred_label, and gold evidence. To create knowledge conflicts, we use word_counter_evidence and sentence_counter_evidence to represent counter-entity and counter-semantic content, respectively.
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 5 January 2026
- Document type
- Dataset