Reproduce “Reason to Retrieve: Enhancing Query Understanding through Decomposition and Interpretation” 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
Yunfei Zhong, Jun Yang, Yixing Fan, Lixin Su, Maarten de Rijke, Ruqing Zhang, Xueqi Cheng
Researchers from ICT, CAS, Baidu Inc., and the University of Amsterdam developed ReDI, a unified framework that enhances query understanding for complex, long-form queries by leveraging LLMs for query decomposition and adaptive interpretation. The system achieved the highest average nDCG@10 of 30.8% for sparse retrieval on the BRIGHT benchmark and demonstrated strong out-of-domain generalization with an average nDCG@10 of 44.9% on BEIR, while also introducing the Coin dataset for complex query research.
Key insights
- Query decomposition, when combined with adaptive interpretations, is highly effective for complex, reasoning-intensive queries, provided the application method is properly designed.
- Explicit reasoning and decomposition (ReDI) significantly enhances lightweight retrieval systems (BM25, SBERT) but can be less synergistic with intrinsically reasoning-native retrievers (e.g., ReasonIR).
- Distilling advanced query understanding from large foundation models into smaller, efficient models is feasible, enabling practical deployment in real-world search environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.