Knowledge graph

A knowledge graph is a knowledge base that represents information as a graph of entities and the relationships between them, layered with a schema that gives the data machine-readable meaning. A node represents a thing, such as a person, place, concept, or document. An edge represents how two things relate, such as born_in, authored, or subclass_of. Because the relationships are first-class records rather than implicit joins, a knowledge graph captures how things connect as directly as it captures the things themselves.

The graph is semantic in the sense that its edge types and node types are drawn from a shared vocabulary, an ontology, so any consumer that understands the vocabulary understands the data. This is what separates a knowledge graph from a plain graph of nodes and edges. The schema lets a reader, or a reasoner, infer new facts from existing ones, eg. that a thing of type Novel is also a Work, or that spouse_of is symmetric.

The RDF model

Knowledge graphs are most often built on the W3C Resource Description Framework (RDF), in which every fact is a triple: a subject, a predicate, and an object. The subject and predicate are IRIs, globally identifiable, and the object is either another IRI, a node, or a literal value. A knowledge graph is then the union of its triples, a labeled directed graph. Layers above RDF, namely RDF Schema and the Web Ontology Language (OWL), add class hierarchies, property domains and ranges, and inference rules, so the graph can be reasoned over rather than only looked up.

Triples are queried with SPARQL, a query language that matches graph patterns against the store. They are serialized for interchange in formats such as Turtle and RDF/XML, or as JSON-LD for the web. Graphs and their data are exchanged between tools using GraphML.

Knowledge graph vs graph database

The terms are often conflated but name different things. A graph database is a storage technology, a database that keeps nodes and edges as first-class records. A knowledge graph is the data and the model: the entities, the relationships, and the schema that describes them. A knowledge graph is frequently stored and queried in a graph database, typically an RDF triple store, but it could equally live in a property-graph store or be exported as a set of files. The graph database is the engine. The knowledge graph is what it holds.

Examples and uses

The best-known instance is Google’s Knowledge Graph, which grounds entities mentioned in a search query against a shared graph and underpins the knowledge panels shown alongside results. Open knowledge graphs such as Wikidata and DBpedia publish entities and relations from Wikipedia as RDF, and academic projects such as YAGO link text and structured sources. Enterprises build internal knowledge graphs to unify data scattered across silos.

Common uses rely on the graph’s connected, typed structure.

  • Search enrichment surfaces entities and related facts alongside query results.
  • Question answering resolves a question to entities and traverses the graph to assemble an answer.
  • Recommendation walks similarity and co-occurrence edges to suggest related items.
  • Data integration reconciles entities from independent sources against a shared schema.
  • Entity disambiguation uses typed context to tell one Jane Austen from another.

See also

References