GraphRAG

Microsoft method

A retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from. method from Microsoft Research designed for questions about a corpus as a whole. A language modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. reads every chunk and records the people, organizations, places, events and other entities it mentions, with the relations among them, in a knowledge graphknowledge graph, n.: A huge web of facts stored as connections between things — 'Paris is the capital of France', 'Marie Curie won a Nobel Prize' — that computers can search, combine and reason over.; the Leiden clustering algorithm then splits the graph into nested groups of tightly linked entities, and the model writes a report on each group in advance. A broad question, such as what themes run through the collection, is answered map-reduce style, by drafting an answer from each relevant group report and then merging the drafts, while a question about specific things starts from the matching entities and collects their neighbors and source passages. Its developers reported fuller and more varied answers to such corpus-wide questions than ordinary vector RAG, which fetches a few similar chunks, at a much higher indexing cost.

Plain RAG couldn't say what the 4,000 interviews had in common; GraphRAG's global search could.

General

Any retrieval-augmented approach that uses a graph — an existing knowledge graph or one extracted from text — to retrieve connected facts, follow multi-hop relationships or organize context, rather than retrieving independent passages by similarity alone.