Knowledge Graph

A knowledge graph is a graph-structured database where knowledge is stored as entities (nodes) and relationships (edges), with properties attached to both. Unlike relational databases that store data in tables, knowledge graphs model information as interconnected networks, making it natural to represent and query complex relationships between concepts.

Knowledge graphs excel at multi-hop reasoning -- answering questions like "find friends-of-friends who work at competitors" -- which would require expensive self-joins in SQL. They are used by major companies including Google (Google Knowledge Graph), Microsoft (GraphRAG), and Amazon (product knowledge graphs). In hybrid RAG+KG systems, the knowledge graph provides structured facts and relationship traversal while RAG provides unstructured document retrieval.

Common implementations use graph databases like Neo4j with query languages such as Cypher or SPARQL. Knowledge graphs are built from triples (subject-predicate-object), and their construction involves entity extraction, relation extraction, entity linking, and schema design.