Research map / Large language models / Retrieval and knowledge
Syllogism-Inspired TableQA: Evidentialization Makes Decomposition Reasoning and Answer Verification More Reliable
Zhe Zhang, Lili Bai, Chaopeng Guo, Jie Song
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 SIRV instead of TIDE (joint DP&Agent) | TableQA with GPT-3.5 · WTQ | exact match accuracy: 75.00 -> 77.89 | Improved |
| use SIRV instead of TIDE (joint DP&Agent) | TableQA with GPT-3.5 · TabFact | exact match accuracy: 89.82 -> 91.85 | Improved |
| use SIRV-Agent instead of TIDE-Agent | Agent mode · WTQ | exact match accuracy: 68.72 -> 73.96 | Improved |
| use SIRV-DP instead of Norm-DP | DP mode · WTQ | exact match accuracy: 66.99 -> 67.38 | Improved |
| add evidential answer verification (SIRV-AV) (ablation) | SIRV · WTQ joint DP&Agent | exact match accuracy: 74.31 -> 77.89 | Improved |
| add evidence-centered decomposition reasoning (SIRV-DR) (ablation) | SIRV · WTQ Agent mode | exact match accuracy: 68.01 -> 73.96 | Improved |
| add evidence-centered decomposition reasoning (SIRV-DR) (ablation) | SIRV · TabFact joint DP&Agent | exact match accuracy: 87.26 -> 91.85 | Improved |
| add SIRV-DR and SIRV-AV to Agent mode (vs standard CoT) (ablation) | SIRV-Agent · WTQ joint | exact match accuracy: 71.91 -> 77.89 | Improved |
| use SIRV-DR/AV only in DP while Agent uses standard CoT (ablation) | SIRV · WTQ Agent mode | exact match accuracy: 67.38 vs 67.38 (DP unchanged) | No clear effect |
| replace self-correction prompting with evidential verification (ablation) | SIRV-AV · WTQ and TabFact | answer correction rate: 47.39 -> 55.71 (mean) | Improved |
| increase candidate answers from 1 to 5 in each mode (ablation) | SIRV majority voting · WTQ | exact match accuracy: 70.07 -> 77.89 | Improved |
| use 5 DP and 5 Agent answers instead of 3:5 on TabFact (ablation) | SIRV majority voting · TabFact | exact match accuracy: 91.85 (3:5) vs 91.21 (5:5) | Hurt |
Closest papers by meaning
- TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · similarity 0.87)
- Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding (ICLR 2024 · similarity 0.86)
- Accurate and Regret-Aware Numerical Problem Solver for Tabular Question Answering (AAAI 2025 · similarity 0.86)
- CABINET: Content Relevance-based Noise Reduction for Table Question Answering (ICLR 2024 · similarity 0.86)
- ORACLE: Optimizing Reasoning Abilities of Large Language Models via Constraint-Led Synthetic Data Elicitation (AAAI 2026 · similarity 0.85)
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 2025 · similarity 0.85)
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.
