---
title: 'Knowledge graph'
description: 'A graph-shaped representation of entities or concepts and their named relationships, often governed by a schema and linked to supporting sources.'
canonical_url: 'https://darkfactory.dev/glossary/knowledge-graph'
markdown_url: 'https://darkfactory.dev/glossary/knowledge-graph.md'
collection: glossary
date_published: '2026-08-04T00:00:00-04:00'
date_modified: '2026-08-26T00:00:00-04:00'
---

# Knowledge graph


## Definition

A knowledge graph is a graph-shaped representation of entities or concepts and their named relationships. Nodes may represent people, products, documents, places, events, or abstract concepts. Edges represent claims such as "works for," "depends on," "located in," or "supersedes." Types, identifiers, constraints, and an ontology or schema give those relationships consistent meaning.

A useful graph separates a claim from its evidence. Provenance can record which document, database row, observation, or extractor supports an edge, along with time and confidence. The graph may then support structured queries, neighborhood exploration, path finding, entity resolution, or rule-based inference. Missing edges do not necessarily mean a relationship is false; the graph may simply be incomplete.

## Where the name became prominent

Graph-based knowledge representation predates the current term by decades. Google's 2012 launch of its capitalized Knowledge Graph product helped popularize the modern name. Amit Singhal described a search model centered on real-world entities and their relationships, summarized as "things, not strings." The announcement is evidence for a prominent product and public usage, not a claim that Google invented knowledge graphs in general.

## Distinguish it from nearby terms

- A **database graph** stores nodes and edges. It becomes a knowledge graph when those elements represent a knowledge domain with meaningful identities and relations.
- An **ontology** defines concepts, relation types, and constraints. A knowledge graph contains instance data and may use an ontology.
- A **vector index** retrieves by learned similarity. A knowledge graph follows explicit relationships. Many systems combine them.
- A **control graph** represents permitted workflow transitions. It is not a knowledge graph merely because both use nodes and edges.

## Operational significance

Decide which artifact is authoritative. In many AI systems, the graph should be a rebuildable projection over source material, especially when a model extracted it. Preserve stable identifiers and source links, version the schema, record extraction methods, and define how corrections propagate. Evaluate entity resolution, relation precision and recall, freshness, and the accuracy of queries that depend on several edges.

## Check your understanding

A graph states that a policy supersedes an older policy. A trustworthy answer should expose both source documents and the effective date, not rely on the edge label alone.

## Related terms

- [GraphRAG](https://darkfactory.dev/glossary/graphrag)
- [Semantic search](https://darkfactory.dev/glossary/semantic-search)
- [Vector database](https://darkfactory.dev/glossary/vector-database)
- [Provenance](https://darkfactory.dev/glossary/provenance)
- [Graph engineering](https://darkfactory.dev/glossary/graph-engineering)

## Related factory areas

- [Context, memory, knowledge & skills](https://darkfactory.dev/factory/context-memory-skills)

## Evidence and further reading

- [Introducing the Knowledge Graph: Things, Not Strings](https://blog.google/products-and-platforms/products/search/introducing-knowledge-graph-things-not/)
- [Microsoft GraphRAG Documentation](https://microsoft.github.io/graphrag/)
- [From Local to Global: A Graph RAG Approach to Query-Focused Summarization](https://arxiv.org/abs/2404.16130)
- [Turing Post: Is Graph Engineering Real?](https://www.turingpost.com/p/is-graph-engineering-real-why-everyone-is-talking-about-it)