AI Glossary
Knowledge graph
knowledge graph, knowledge graph, GraphRAG
A knowledge graph is an organized network of entities (people, products, documents) and the relationships between them. It can ground a model's answers — either complementing or replacing vector search.
- It represents knowledge as entities joined by named relationships, not as loose chunks of text.
- In the GraphRAG variant, the model answers from paths through the graph rather than from the nearest vectors alone.
- It handles questions well when they require connecting facts across several documents.
A knowledge graph is a way of recording knowledge as a network: nodes are entities (for example, a customer, a contract, a product), and edges are named relationships between them (for example, "the customer signed the contract"). Unlike a document store, where information sits buried in the text, a graph makes the connections between facts explicit and searchable.
In the context of language models, a knowledge graph can be the foundation of an approach known as GraphRAG. Instead of retrieving the nearest chunks from a vector database — as in classic RAG — the system traverses the graph's relationships and hands the model related facts. This helps with questions that require pulling together information scattered across many sources, where semantic search alone returns relevant but disconnected fragments.
From a deployment standpoint, a graph and vector search are not mutually exclusive — they often work together. Vectors are good at finding "what a text is about," while a graph is good at answering "how these things connect." The cost of a graph is in building and maintaining it: someone has to define which entities and relationships matter, and keep them consistent as the company's data changes.
Related terms
Related articles