Reproduce “DMQR-RAG: Diverse Multi-Query Rewriting for RAG” 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
Zhicong Li, Jiahao Wang, Zhishu Jiang, Hangyu Mao, Zhongxia Chen, Jiazhen Du, Yuanxing Zhang, Fuzheng Zhang and 2 more
DMQR-RAG is a framework for Retrieval-Augmented Generation systems that addresses the query rewriting bottleneck by systematically creating diverse queries based on different levels of information content. This approach, coupled with an adaptive selection mechanism, achieves up to a 14.46% improvement in Precision@5 on academic benchmarks and a 10.0% increase in Precision@5 in industrial settings.
Key insights
- Pseudo-Answer Rewriting (PAR) is the most impactful strategy, with its removal causing the largest performance decrease in ablation studies.
- Using all rewriting strategies together is not always optimal; an adaptive selection mechanism is crucial for balancing diversity with efficiency and performance.
- Adaptive strategy selection reduced the average number of rewrites by nearly 40% (from 4 to ~2.5), leading to efficiency gains and consistent performance improvements, particularly for smaller LLMs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.