Research map / Reinforcement learning / RL from feedback
Repurposing Synthetic Data for Fine-grained Search Agent Supervision
Yida Zhao, Kuan Li, Xixi Wu, Liwen Zhang, Ding-Chu Zhang, Baixuan Li, Maojia Song, Zhuo Chen, Chenxi Wang, Xinyu Wang, Kewei Tu, Pengjun Xie and 3 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 |
|---|---|---|---|
| replace outcome-only reward with entity-aware reward (partial credit for wrong rollouts) | GRPO, Qwen2.5-7B-Instruct, Local env · QA benchmarks avg | Pass@1: 61.4 -> 64.2 | Improved |
| entity-aware reward, evaluated in Web env | GRPO, Qwen2.5-7B-Instruct · QA benchmarks avg (Web) | Pass@1: 66.2 -> 67.8 | Improved |
| entity-aware reward on deep research tasks | GRPO, 7B · GAIA | Pass@3: 44.7 -> 51.5 | Improved |
| entity-aware reward at 30B scale | GRPO, Qwen3-30B-A3B · BrowseComp-ZH | Pass@3: 38.8 -> 41.2 | Improved |
| entity-aware reward | GRPO, 7B · BrowseComp | Pass@1: 6.3 -> 9.3 | Improved |
| entity-aware reward | GRPO, 30B · GAIA | Pass@1: 47.6 -> 48.5 | Improved |
| entity-aware reward | GRPO, 7B Local env · PopQA | Pass@1: 50.4 -> 50.2 | No clear effect |
| entity-aware reward | GRPO, 7B Web eval · HotpotQA | Pass@1: 73.8 -> 73.7 | No clear effect |
| entity-aware reward | GRPO · tool calls per rollout | tool calls | Improved |
| entity-aware reward | GRPO, Web-30B · training accuracy | training accuracy | Improved |
| increase entity weight alpha from 0.0 to 0.3 (ablation) | E-GRPO · four deep research benchmarks | accuracy | Improved |
| increase entity weight alpha from 0.3 to 0.5 (ablation) | E-GRPO · four deep research benchmarks | accuracy | Hurt |
Closest papers by meaning
- ReSeek: A Self-Correcting Framework for Search Agents with Instructive Rewards (ICML 2026 · similarity 0.88)
- Empowering Efficiency and Efficacy in WebAgent via Enabling Info-Rich Seeking (ICLR 2026 · similarity 0.88)
- Search More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization (NeurIPS 2026 · similarity 0.88)
- Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers (NeurIPS 2025 · similarity 0.87)
- GISA: A Benchmark for General Information-Seeking Assistant (NeurIPS 2026 · similarity 0.86)
- HierSearch: A Hierarchical Enterprise Deep Search Framework Integrating Local and Web Searches (AAAI 2026 · 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.
