Research map / Large language models / Retrieval and knowledge
AVeriTeC: A Dataset for Real-world Claim Verification with Evidence from the Web
Michael Schlichtkrull, Zhijiang Guo, Andreas Vlachos
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use gold question-answer evidence instead of retrieved evidence | AVeriTeC baseline · AVeriTeC test | veracity accuracy at λ=.25: 0.15 -> 0.49 | Improved |
| Use gold question-answer evidence instead of retrieved evidence | AVeriTeC baseline · AVeriTeC test | justification METEOR at λ=.25: 0.07 -> 0.28 | Improved |
| Use no search or evidence (ablation) | AVeriTeC baseline · AVeriTeC test | veracity accuracy at λ=.25: 0.15 -> 0.02 | Hurt |
| Use no search or evidence (ablation) | AVeriTeC baseline · AVeriTeC test | veracity macro-F1: .23 -> .17 | Hurt |
| Generate one ChatGPT paraphrase per claim and include it in stance-model training (ablation) | BERT-large stance detection · AVeriTeC gold evidence | veracity macro-F1: .49 -> .46 | Hurt |
| Add predicted veracity as an input to justification generation (ablation) | BART-large justification generation · AVeriTeC | justification METEOR | No clear effect |
| Use BLOOM instead of Vicuna for generating search questions | AVeriTeC search-question generation · AVeriTeC | retrieval performance | Improved |
| Use 10 retrieved in-context examples for search-question generation | BLOOM search-question generation · AVeriTeC | retrieval performance | Improved |
| Include the claim when generating questions from evidence passages (ablation) | BLOOM passage-question generation · AVeriTeC | question-generation performance | Hurt |
| Use 10 retrieved in-context examples for passage-question generation | BLOOM passage-question generation · AVeriTeC | question-generation performance | Improved |
| Use BERT-large rather than BLOOM-7b1 or Vicuna-13b for veracity prediction with gold evidence | AVeriTeC veracity prediction · AVeriTeC gold evidence | veracity macro-F1: .43, .48 -> .49 | Improved |
| Use BART-large rather than BLOOM-7b1 or Vicuna-13b for justification generation | AVeriTeC justification generation · AVeriTeC | METEOR: .28, .23, and .25 | Improved |
Closest papers by meaning
- AVerImaTeC: A Dataset for Automatic Verification of Image-Text Claims with Evidence from the Web (NeurIPS 2025 · similarity 0.92)
- Heterogeneous Graph Reasoning for Fact Checking over Texts and Tables (AAAI 2024 · similarity 0.86)
- DEFAME: Dynamic Evidence-based FAct-checking with Multimodal Experts (ICML 2025 · similarity 0.85)
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 2025 · similarity 0.85)
- MRR-FV: Unlocking Complex Fact Verification with Multi-Hop Retrieval and Reasoning (AAAI 2025 · similarity 0.85)
- Beyond Text: Fine-Grained Multi-Modal Fact Verification with Hypergraph Transformers (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
