Reproduce “SimGRAG: Leveraging Similar Subgraphs for Knowledge Graphs Driven Retrieval-Augmented Generation” 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

Yuzheng Cai, Zhenyue Guo, Yiwen Pei, Wanrui Bian, Weiguo Zheng

SimGRAG aligns KG retrieval with natural-language questions by having an LLM turn each query into a graph pattern, then retrieving structurally matching KG subgraphs ranked by semantic distance. Retrieved triples are passed to an LLM to answer the question. On MetaQA’s three-hop movie task with Llama 3 70B, it reaches 97.8 Hits@1 versus 88.2 for KG-GPT†, which receives oracle entities. The approach assumes understandable KG schemas.

Key insights

  • Combining structural constraints (graph isomorphism) with semantic similarity (GSD) effectively bridges the gap between natural language queries and structured knowledge for precise subgraph retrieval.
  • Leveraging LLMs for query-to-pattern conversion, guided by few-shot examples, enables plug-and-play adaptability on human-understandable KGs without requiring specific model training.
  • An optimized retrieval algorithm, incorporating pre-filtering and dynamic pruning during graph traversal, is crucial for achieving efficient and scalable subgraph retrieval from massive knowledge graphs.

Summary by alphaXiv.

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