A knowledge graph is a structured representation of knowledge in which entities (people, places, things, concepts) are represented as nodes and their relationships as edges. It uses identifiers and shared semantics (often through RDF and OWL) so that different datasets can be connected, queried (SPARQL), and reasoned over consistently.
Key ideas
- Semantics first: data is described using vocabularies/ontologies (classes, properties, constraints).
- Graph-oriented: knowledge = nodes (entities) + edges (relationships) + optional labels or weights.
- Interoperability: global identifiers (IRIs) and shared schemas make it possible to connect diverse sources.
- Querying and reasoning: graph patterns via SPARQL; inference using RDFS/OWL rules.
- Evolution: graphs can grow incrementally without rigid table migrations.
Typical components
- Data model: RDF triples/quads or property graphs; mappings from relational data/JSON.
- Vocabulary: classes and properties (e.g. schema.org, FOAF, SKOS) and custom ontologies.
- Storage and services: triplestores / graph databases, SPARQL endpoints, APIs.
- Governance: provenance (PROV), versioning, data quality, access control.
Use cases
- Search and recommendations: entity understanding, disambiguation, and connection of related entities.
- Enterprise data integration: unifying silos through a common semantic layer.
- Analysis and reasoning: impact analysis, compliance checks, and root-cause paths.
- Content and media: thematic maps, editorial linking, rights and metadata management.
In applied artificial intelligence systems, knowledge graphs can act as a layer that connects data, entities, context, and criteria within a broader processing and automation architecture.
Mini example
This type of structure allows a system not only to store data, but also to work with the relationships between data elements as part of its context.
Triples (subject–predicate–object):
<http://example.org/person/123> a schema:Person .
<http://example.org/person/123> schema:name "Ada Lovelace" .
<http://example.org/person/123> schema:knows <http://example.org/person/456> .Query to find who Ada knows (SPARQL):
SELECT ?friendName WHERE {
<http://example.org/person/123> schema:knows ?friend .
?friend schema:name ?friendName .
}Related terms
- RDF (Resource Description Framework)
- RDFS and OWL
- SPARQL
- Ontology and SKOS
Frequently asked questions
Is a knowledge graph the same as a graph database?
Not necessarily. A knowledge graph is a model + meaning (semantics and governance). It can be stored in a triplestore or a property graph database, but the key point is that nodes and edges have explicit semantic definitions.
Do I need RDF?
RDF is the most widely used standard for semantic interoperability. Some projects begin with property graphs and later map them to RDF in order to link and publish data.
More information
- schema.org (basic web vocabulary)
- W3C specifications for RDF, RDFS, OWL and SPARQL
To see how this logic of connecting data, structure, and criteria can become part of an applied AI system, read AI Systems: Data, Code and Automation Beyond the Chat.
« Back to Glossary Index
