Reproduce “QREF: Improving DeepSearch Agents by Aligning Search with Reasoning via Query Refinement and Evidence Filtering” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yujia Chen, Yuhui Wang, Junzhe Wang, Shihan Dou, Di Liang, Yujiong Shen, Yuhao Zhou, Zhiheng Xi and 10 more

QREF improves web-search agents by refining query batches and filtering retrieved snippets using the question and the agent’s current reasoning. Its refiner is trained with supervised examples and reinforcement learning based on entity-level retrieval outcomes. On WebSailor-32B’s BrowseComp evaluation, RL-trained QREF raises Pass@1 from 10.82% to 17.61%, while average search turns rise from 6.75 to 7.77.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.