← AI Terminology

Knowledge Graph

A knowledge graph represents facts as a network of entities (nodes) and relationships (edges), often with types and properties, enabling structured queries and multi-hop traversal.

Used in search, RAG, enterprise data fabrics, and classic semantic web stacks.
Why It Matters in AI
Unstructured text hides relationships; graphs make them queryable and auditable. Combined with LLMs for extraction and NL2query, graphs ground answers in explicit structure — key for compliance-heavy domains.
Key Points
Aspect Description
Model Entities, relations, attributes; RDF/property graphs
Query Cypher, SPARQL, Gremlin
Stores Neo4j, Amazon Neptune, RDF triple stores
AI role LLM entity/relation extraction; graph-enhanced RAG
Benefit Explainable paths; deduplicated entities
Related GraphRAG, ontologies, link prediction
Simple Analogy
A detective’s corkboard: people and companies as pins, strings for relationships you can follow with your finger.
Common Usage Examples
  • Neo4j Cypher: MATCH (a)-[r]->(b)
  • LLM extract triples into a graph
  • GraphRAG community detection
  • Enterprise “customer 360” graphs
Summary
In short: A knowledge graph stores facts as linked entities and relations — structured memory that LLMs can query and explain.