Reproduce “Knowledge Graph RAG: Agentic Crawling and Graph Construction in Enterprise Documents” 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

Koushik Chakraborty, Koyel Guha

An Agentic Knowledge Graph (AKG) framework featuring recursive reference crawling enhances Retrieval-Augmented Generation (RAG) by integrating deterministic graph traversal for complex enterprise documents. This approach, which models temporal and hierarchical relationships, achieved a 70% accuracy improvement over standard vector-based RAG on the Code of Federal Regulations dataset.

Key insights

  • Explicitly modeling temporal and hierarchical relationships through a knowledge graph and structured edges significantly improves the accuracy and completeness of information retrieval in complex document sets.
  • Deterministic graph traversal, guided by an autonomous agent, effectively resolves multi-hop dependencies and superseding logic, mitigating issues like temporal hallucination.
  • Hybrid RAG architectures that combine semantic search with rule-based, graph-driven navigation can overcome the limitations of purely probabilistic retrieval methods in highly structured domains.

Summary by alphaXiv.

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