Graph structure can help a system connect a customer, a contract, and the products covered by that contract. It can also support questions that require synthesizing information across documents rather than finding one similar passage.
Microsoft’s GraphRAG builds an entity graph and community summaries from source text. Its global search uses those summaries to address broader corpus questions, while local search works around particular entities. Other graph-based RAG systems may query an existing graph or combine it with vector retrieval.
Example: asking “Which services depend on this component?” benefits from explicit dependency links. But an outdated or wrongly extracted link can mislead the answer. A graph is structured evidence, not an automatic guarantee of truth.
Compare against the simplest retrieval baseline that serves your questions. Include extraction, indexing, update cost, and answer quality. Vector database vs. knowledge graph explores the architecture choice; the right choice depends on the relationships the task needs.
Sources
- Microsoft: GraphRAG documentation — Explains the specific indexing pipeline, community summaries, and global/local query modes.
Go deeper
- Vector database or knowledge graph — which should you use for AI retrieval? AI, decoded · Vector Database vs. Knowledge Graph for AI Retrieval
- What is RAG, and why do AI systems use it? AI, decoded · What Is RAG (Retrieval-Augmented Generation)
- What is agentic RAG, and how is it different from regular RAG? AI, decoded · Agentic RAG vs. Traditional RAG
- Microsoft GraphRAG: Getting started docs
Follow the sample from indexing to local and global questions. Running it needs Python and configured model access; indexing can consume substantial LLM resources and incur charges.
