What is Knowledge Graph?
Knowledge Graph is a connected representation of entities, relationships, and claims, often with explicit identifiers, provenance, and domain semantics. It may use RDF triples, a property graph, or another storage model; it is not synonymous with a graph database.
Quick Facts
| Created | Concept from 2000s, popularized by Google in 2012 |
|---|---|
| Specification | Official Specification |
How It Works
Knowledge graphs organize facts as subject-predicate-object triples (for example, 'Paris' - 'is capital of' - 'France') or property-graph nodes and edges. They are useful when applications need explicit relationship traversal, provenance, ontology rules, or connected evidence. They do not automatically make an LLM factual: source quality, entity resolution, retrieval authorization, and answer evaluation still determine reliability. For AI retrieval, compare graph traversal and GraphRAG against relational, vector, or hybrid baselines using the <a href="https://qubittool.com/blog/rag-evaluation-production-guide">production RAG evaluation guide</a>.
Key Characteristics
- Represents entities and relationships as graph structures
- Uses subject-predicate-object triples for facts
- Can support SPARQL, Cypher, or application-specific graph queries
- Can apply explicit ontology or rule semantics when configured
- Preserves source and revision lineage for graph claims
- Can augment LLM retrieval when it outperforms simpler baselines
Common Use Cases
- Search engine knowledge panels
- Recommendation systems with relationship awareness
- Question answering with factual grounding
- Drug discovery and biomedical research
- Enterprise knowledge management
Example
Loading code...Frequently Asked Questions
What is the difference between a knowledge graph and a database?
A knowledge graph is a representation of entities, relationships, claims, and semantics; a database is a storage and query system. A knowledge graph may be stored in an RDF store, property-graph database, relational database, or hybrid indexes. Choose the implementation from query shape, transactions, semantics, authorization, and operational constraints.
How do knowledge graphs reduce AI hallucinations?
Knowledge graphs can supply structured evidence and provenance to an LLM, but they do not guarantee factual answers. Reliability still depends on source quality, retrieval coverage, authorization, prompt design, and whether generated claims are supported by the retrieved evidence.
What is a triple in knowledge graph terminology?
In RDF, a triple is a subject-predicate-object statement, such as 'Paris' - 'isCapitalOf' - 'France'. The subject and predicate are IRIs, while the object may be an IRI, literal, or blank node. A collection of RDF triples forms an RDF graph; property graphs use a different node, relationship, and property model.
What is SPARQL and how is it used with knowledge graphs?
SPARQL is the W3C query language for RDF data. It matches graph patterns, filters and aggregates results, and can query RDF datasets containing a default graph and named graphs. It applies to RDF; property-graph systems commonly use other languages such as Cypher.
What are some popular knowledge graphs I can use?
Public knowledge graphs include Wikidata (structured Wikipedia data), DBpedia (extracted from Wikipedia), and Google's Knowledge Graph (powers search panels). For building custom graphs, popular tools include Neo4j (property graph database), Apache Jena (RDF framework), and Amazon Neptune (managed graph database).