Reproduce “When RAG Meets Query Planning: Logical Query Trees for Resolving Exploratory Reasoning Problems” 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

Ganlin Xu, Linghao Zhang, Zhitao Yin, Hongda Xi, Chen Yang, Jiaqing Liang, Weijia Lu, Sihang Jiang and 2 more

PlanRAG introduces a framework that leverages database-style query planning to enhance Retrieval-Augmented Generation (RAG) systems in resolving complex, uncertain "Exploratory Reasoning Problems" (ERPs) by constructing and optimizing Logical Query Trees (LQTs) to coordinate retrieval and generation. It achieved superior performance on a new WikiWeb-ERP dataset and improved efficiency through parallel execution.

Key insights

  • Adapting database query planning principles, specifically Logical Query Trees and dynamic programming with a tailored cost model, can effectively structure and optimize retrieval-augmented generation for complex natural language queries.
  • Explicitly performing global structural optimization before execution significantly reduces retrieval noise and error accumulation in RAG, addressing limitations of sequential or loosely structured reasoning.
  • A multi-dimensional cost model incorporating semantic similarity, structural density, depth, and balance is crucial for generating high-quality, robust, and semantically aligned LQTs from ambiguous natural language.

Summary by alphaXiv.

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