A knowledge graph is a way of organizing information as a network of entities (the things in the world, such as people, places, products, or concepts) and the relationships that connect them. Instead of storing facts in isolated tables or documents, a knowledge graph links them together so that a statement like "Paris is the capital of France" is represented as a structured triple: a subject (Paris), a predicate (is the capital of), and an object (France). This structure lets software traverse connections, follow chains of meaning, and surface answers that go beyond keyword matching.
How a knowledge graph works
At its core, a knowledge graph is built from triples expressed in a formal schema, most commonly the Resource Description Framework (RDF) or the property-graph model used by databases such as Neo4j. Each triple asserts a fact: (Marie Curie) — discovered — (Radium). Nodes carry properties and types (e.g., a "Person"), while edges carry the nature of the relationship. A schema or ontology defines what kinds of entities and relations are allowed, which lets the graph enforce consistency and support basic reasoning.
Construction is typically a pipeline of extraction and integration. Source documents, databases, or web pages are parsed, named entities are recognized, candidate relationships are extracted (often with machine-learning models), and the results are resolved against an existing graph to merge duplicates. Query languages like SPARQL or Cypher then let applications ask graph-shaped questions, such as "which scientists worked at institutions funded by X and published on Y?" — something that would require costly joins in a relational database.
Why it matters
Knowledge graphs give AI systems a shared, explicit substrate of facts to draw on. Search engines use them to power direct-answer panels; recommendation systems use them to find related items through shared attributes; and large language models use them as a source of grounded, up-to-date information through techniques such as retrieval-augmented generation. By making relationships first-class, a knowledge graph also makes provenance and context traceable, which is critical in domains like healthcare, finance, and enterprise knowledge management where hallucination and stale data are real risks.
Key types
- Open / public knowledge graphs — large, general-purpose graphs such as Wikidata, DBpedia, and Google's Knowledge Graph, built from public sources and used to enrich search and assistants.
- Enterprise knowledge graphs — private graphs that unify a company's internal data (customers, products, contracts, assets) for analytics, compliance, and AI applications.
- Domain knowledge graphs — focused graphs in a specific field, such as biomedical (e.g., UMLS, Gene Ontology) or materials science, where vocabulary control and curation matter more than breadth.
- Multimodal knowledge graphs — extensions that link text nodes to images, video, or audio, enabling cross-modal retrieval and reasoning.
For all its variants, a knowledge graph's defining feature is that relationships are as queryable as the things they connect — turning scattered facts into a navigable, machine-readable map of a domain.
