DEEP-PolyU/Awesome-GraphRAG

Organizes GraphRAG research into three core dimensions—knowledge organization (graph construction via entity extraction or hierarchical indexing), retrieval mechanisms (semantic similarity, logical reasoning, GNN-based, and LLM-based approaches), and knowledge integration (fine-tuning vs. in-context learning)—with accompanying benchmarks and open-source implementations. The repository maps distinct GraphRAG paradigms including knowledge-based approaches that extract entity-relation graphs from raw text and index-based approaches that build topic hierarchies, contrasting both against traditional chunk-based RAG. Covers integration patterns across LLM frameworks and provides curated links to peer-reviewed papers, established benchmarks like GraphRAG-Bench, and reference implementations including LinearRAG and LogicRAG.

2,181 stars.