Academic publications, surveys, and benchmark studies
Survey of methods for using LLMs to automatically construct knowledge graphs.
A comprehensive benchmark for evaluating GraphRAG systems across multiple domains.
Novel approach improving GraphRAG entity extraction through optimized prompt engineering and structured output.
Systematic comparison showing GraphRAG achieves up to 86.31% accuracy on RobustQA benchmarks, a 3x improvement.
The foundational GraphRAG paper by Microsoft Research introducing the complete pipeline with Leiden clustering and community summaries.
Uses GNNs to perform reasoning over knowledge graphs before passing context to LLMs.
Comprehensive survey categorizing GraphRAG approaches into graph-based indexing, graph-guided retrieval, and graph-enhanced generation.
Framework for faithful LLM reasoning with planning-retrieval-reasoning decomposition.
LLMs perform step-by-step reasoning by traversing knowledge graph paths.
Comprehensive roadmap for integrating LLMs with knowledge graphs.