Reproduce “KET-RAG: A Cost-Efficient Multi-Granular Indexing Framework for Graph-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

Yiqian Huang, Shiqi Zhang, Xiaokui Xiao

Researchers at the National University of Singapore developed KET-RAG, a multi-granular indexing framework for Retrieval-Augmented Generation (RAG) that significantly reduces the indexing cost of Graph-RAG systems. This framework achieves comparable or superior retrieval and generation quality to state-of-the-art methods while decreasing indexing costs by over an order of magnitude.

Key insights

  • Identifying and processing only a small fraction of structurally important 'core chunks' with LLMs for knowledge graph construction can drastically reduce indexing costs without substantial performance degradation.
  • A text-keyword bipartite graph, constructed without intensive LLM processing, offers an effective and lightweight alternative for capturing in-text relationships, outperforming vanilla text retrieval.
  • The hybrid approach, combining a sparse, high-fidelity knowledge graph skeleton with a comprehensive, cost-efficient text-keyword bipartite graph, provides a superior balance of accuracy and cost compared to monolithic Graph-RAG designs.

Summary by alphaXiv.

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