FaceNews V2.0: A Thai Fake News Detection Dataset

FaceNews V2.0 is a Thai-language fake news detection dataset containing 25,623 metadata records compiled from multiple fact-checking platforms and verified news sources in Thailand. This public release is metadata-only and includes id, label, source, url, category, published_at, and split fields. Full article text is not publicly redistributed due to copyright restrictions on commercial news publishers. Labels are derived from source-provided fact-check verdicts for fact-checking sources and editorial-source credibility for established news outlets. AI-generated labels are not used as ground truth. Sources: AFNC (Anti-Fake News Center, data.go.th), Cofact Thailand, Thai PBS Verify, Matichon, Khaosod, and The Standard. Label distribution: FAKE 40.4% / REAL 38.6% / UNVERIFIED 15.3% / MISLEADING 5.8%. Splits: Train 20,491 / Val 2,563 / Test 2,569 (stratified, seed=42). Research-use notes for reconstructing article text from URLs where permitted are provided in scrape_instructions.md. Users are responsible for complying with each source's terms of service, robots.txt directives, and applicable copyright law. V2.0.1 is a documentation-only update. No records were added, removed, relabeled, or reordered. The metadata CSV is unchanged from V2.0. Intended uses: Thai NLP benchmarking, misinformation detection, cross-lingual transfer learning to ASEAN languages.