Research map / Large language models / Retrieval and knowledge
TopBench: A Benchmark for Implicit Predictive Reasoning in Tabular Question Answering
An-Yang Ji, Jun-Peng Jiang, De-Chuan Zhan, Han-Jia Ye
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 |
|---|---|---|---|
| add semantic hints (target, task type, feature descriptions) (ablation) | Qwen3-Instruct agentic · Single-Point | accuracy: 0.43 -> 0.56 | Improved |
| add semantic hints (ablation) | DeepSeek-V3.2 agentic · Treatment Effect | trend accuracy: 0.57 -> 0.68 | Improved |
| add semantic hints (ablation) | GPT-5.2 agentic · Single-Point | accuracy: 0.60 -> 0.64 | Improved |
| add semantic hints (ablation) | GPT-5.2 agentic · Ranking and Filtering | F1: 0.38 -> 0.48 | Improved |
| add semantic hints (ablation) | Gemini 3 Flash agentic · Decision Making | decision accuracy: 0.65 -> 0.61 | Hurt |
| add semantic hints (ablation) | GPT-5.2 agentic · Decision Making | decision accuracy: 0.55 -> 0.53 | Hurt |
| add semantic hints (ablation) | Gemini 3 Flash agentic · Ranking and Filtering | F1: 0.58 -> 0.57 | No clear effect |
| replace LLM-only agent with predict-only tabular ensemble given gold profile (ablation) | Gemini 3 Flash agentic · Single-Point | accuracy: 0.66 -> 0.76 | Improved |
| replace LLM-only agent with predict-only tabular ensemble given gold profile (ablation) | Gemini 3 Flash agentic · Decision Making | decision accuracy: 0.65 -> 0.72 | Improved |
| replace LLM-only agent with predict-only tabular ensemble given gold profile (ablation) | Gemini 3 Flash agentic · Treatment Effect | trend accuracy: 0.65 -> 0.69 | Improved |
| use agentic code execution instead of text-based reasoning | GPT-5.2 · Treatment Effect | trend accuracy: 0.51 -> 0.65 | Improved |
| use agentic code execution instead of text-based reasoning | Qwen3-Instruct · Single-Point | accuracy: 0.57 -> 0.43 | Hurt |
Closest papers by meaning
- TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · similarity 0.91)
- Accurate and Regret-Aware Numerical Problem Solver for Tabular Question Answering (AAAI 2025 · similarity 0.89)
- TACT: Advancing Complex Aggregative Reasoning with Information Extraction Tools (NeurIPS 2024 · similarity 0.88)
- Text2Analysis: A Benchmark of Table Question Answering with Advanced Data Analysis and Unclear Queries (AAAI 2024 · similarity 0.88)
- Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding (ICLR 2024 · similarity 0.88)
- NeedleInATable: Exploring Long-Context Capability of Large Language Models towards Long-Structured Tables (NeurIPS 2025 · similarity 0.87)
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.
