Research map / Large language models / Retrieval and knowledge
Think Then Rewrite: Reasoning Enhanced Query Rewriting for Domain Specific Retrieval
Ang Li, Yufei Shi, Yuxuan Si, Yiquan Wu, Ming Cai, Xu Tan, Yi Wang, Changlong Sun, Xiaozhong Liu, Kun Kuang
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 CMI reward to the recall+format reward (ablation) | TTR · EQUALS | R@10: 43.26 -> 48.62 | Improved |
| add recall reward (ablation) | TTR · EQUALS | R@10: 40.98 -> 48.62 | Improved |
| add format reward (ablation) | TTR · EQUALS | R@10: 46.41 -> 48.62 | Improved |
| use adaptive negative sampling instead of random negatives only (ablation) | TTR CMI reward · EQUALS | R@10: 47.18 -> 48.62 | Improved |
| use CMI reward instead of similarity-based reward (ablation) | TTR · EQUALS | R@10: 46.54 -> 48.62 | Improved |
| add off-policy golden rewrites (ablation) | TTR · EQUALS | R@10: 39.52 -> 48.62 | Improved |
| use constructed golden rewrites instead of raw target documents as off-policy data (ablation) | TTR · EQUALS | R@10: 44.52 -> 48.62 | Improved |
| TTR reward and off-policy design over vanilla GRPO rewriting | Qwen2.5-1.5B · EQUALS | R@10: 34.71 -> 48.62 | Improved |
| TTR over LUFFY (mixed-policy with recall reward only) | Qwen2.5-1.5B · KUAKE-IR | R@10: 42.39 -> 46.49 | Improved |
| swap GRPO for PPO/DAPO-style RL algorithms | query rewriter · EQUALS | R@10: 28.54 vs 27.05 pure retriever | No clear effect |
| SFT on golden rewrites with reasoning | Qwen2.5-1.5B · KUAKE-IR | R@10: 43.24 -> 36.04 | Hurt |
| increase CMI reward weight | TTR · EQUALS and KUAKE-IR | Recall@10 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DAPO — DAPO: An Open-Source LLM Reinforcement Learning System at Scale (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MaFeRw: Query Rewriting with Multi-Aspect Feedbacks for Retrieval-Augmented Large Language Models (AAAI 2025 · similarity 0.89)
- Search and Refine During Think: Facilitating Knowledge Refinement for Improved Retrieval-Augmented Reasoning (NeurIPS 2025 · similarity 0.88)
- Incorporating Self-Rewriting into Large Language Model Reasoning Reinforcement (AAAI 2026 · similarity 0.88)
- ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- DynamicRAG: Leveraging Outputs of Large Language Model as Feedback for Dynamic Reranking in Retrieval-Augmented Generation (NeurIPS 2025 · similarity 0.86)
- RewriteLM: An Instruction-Tuned Large Language Model for Text Rewriting (AAAI 2024 · similarity 0.86)
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.
