Research map / Large language models / Retrieval and knowledge
Harnessing Temporal Databases for Systematic Evaluation of Factual Time-Sensitive Question-Answering in LLMs
Soyeon Kim, Jindong Wang, Xing Xie, Steven Euijong Whang
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 |
|---|---|---|---|
| require correct time references in addition to correct answer (AT vs A) | 8 LLMs on TDBench · TDBench temporal alignment, Wikipedia | accuracy: 56.9 -> 35.2 (average A -> AT) | Hurt |
| add Chain-of-Thought or Time-CoT prompting instead of RAG on multi-hop questions (ablation) | GPT-4 · TDBench multi-hop | accuracy: 67.2 -> 92.4 (CoT), 93.9 (Time-CoT) | Improved |
| add RAG with retrieved Wikipedia passages (ablation) | GPT-4 · TDBench multi-hop | accuracy: 67.2 -> 35.4 | Hurt |
| add RAG with retrieved Wikipedia passages (ablation) | GPT-3.5 · TDBench multi-hop | accuracy: 63.1 -> 71.2 | Improved |
| add CoT or Time-CoT prompting (ablation) | GPT-3.5 · TDBench multi-hop | accuracy: 63.1 -> 39.4 (CoT), 15.2 (Time-CoT) | Hurt |
| add Time-CoT prompting (ablation) | GPT-4o · TDBench multi-hop | accuracy: 79.3 -> 95.5 | Improved |
| Time-CoT versus plain CoT (ablation) | multi-hop prompting · TDBench multi-hop | accuracy | No clear effect |
| add RAG with retrieved Wikipedia passages (ablation) | Gemma2-27B · TDBench multi-hop | accuracy: 77.8 -> 40.9 | Hurt |
| build TSQA pairs from SQL with answers grounded in database instead of manual authoring | TDBench · agreement with manual verification | F1: 0.97 (alignment), 0.88 (reasoning), 0.91 (multi-hop) | Improved |
Closest papers by meaning
- Test of Time: A Benchmark for Evaluating LLMs on Temporal Reasoning (ICLR 2025 · similarity 0.90)
- RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content (NeurIPS 2024 · similarity 0.88)
- TimE: A Multi-level Benchmark for Temporal Reasoning of LLMs in Real-World Scenarios (NeurIPS 2025 · similarity 0.88)
- ListQA: A Benchmark for Evaluating List-Formatted Factual Knowledge Retrieval in Large Language Models (NeurIPS 2026 · similarity 0.88)
- NovelQA: Benchmarking Question Answering on Documents Exceeding 200K Tokens (ICLR 2025 · similarity 0.87)
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 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.
