Research map / Large language models / Retrieval and knowledge
Careful Queries, Credible Results: Teaching RAG Models Advanced Web Search Tools with Reinforcement Learning
Yuqin Dai, Shuo Yang, Guoqing Wang, Yong Deng, Zhanwei Zhang, Jun Yin, Pengyu Zeng, Zhenzhe Ying, Changhua Meng, Can Yi, Yuchen Zhou, Weiqiang Wang and 1 more
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 Source-restricting Reward (SR) to base RL training (ablation) | WebFilter (Qwen2.5-7B-Instruct) · 2Wiki | ACC_R: 55.2 -> 59.7 | Improved |
| add Source-restricting Reward (SR) to base RL training (ablation) | WebFilter (Qwen2.5-7B-Instruct) · TriviaQA | ACC_L: 82.6 -> 86.0 | Improved |
| add Source-restricting Reward (SR) (ablation) | WebFilter · advanced operator usage in queries | fraction of queries with advanced operators: less than 10% -> more than 75% | Improved |
| add Retrieval-precision Reward (RR) on top of SR (ablation) | WebFilter · Bamboogle | ACC_R: 64.6 -> 73.1 | Improved |
| add Retrieval-precision Reward (RR) on top of SR (ablation) | WebFilter · Bamboogle | ACC_L: 65.2 -> 74.3 | Improved |
| add Retrieval-precision Reward (RR) on top of SR (ablation) | WebFilter · HotpotQA | ACC_R: 50.6 -> 55.1 | Improved |
| add Retrieval-precision Reward (RR) on top of SR (ablation) | WebFilter · NQ | ACC_R: 41.4 -> 40.1 | Hurt |
| add Retrieval-precision Reward (RR) on top of SR (ablation) | WebFilter · TriviaQA | ACC_R: 79.0 -> 77.5 | Hurt |
| add Source-restricting Reward (SR) to base (ablation) | WebFilter · Bamboogle | ACC_R: 64.9 -> 64.6 | No clear effect |
| use WebFilter instead of DeepResearcher on MuSiQue | WebFilter · MuSiQue | ACC_R: 27.1 -> 24.5 | Hurt |
| train with RL over web search with advanced operators | WebFilter · HotpotQA | ACC_R: 52.8 -> 55.1 | Improved |
Closest papers by meaning
- RAGGED: Towards Informed Design of Scalable and Stable RAG Systems (ICML 2025 · similarity 0.90)
- Process vs. Outcome Reward: Which is Better for Agentic RAG Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- RAG-DDR: Optimizing Retrieval-Augmented Generation Using Differentiable Data Rewards (ICLR 2025 · similarity 0.89)
- Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation (ICML 2024 · similarity 0.89)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 2025 · similarity 0.89)
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.
