MMOOZZGG.com

Glossary

What is a knowledge graph

In short

A knowledge graph is a database where facts are written as objects and the links between them: a client signed a contract, a contract belongs to a project.

Search and RAG · knowledge graph

MMOOZZGG.com editorsChecked: 1 min read

For a child

Remember the family tree you draw at school: mom, dad, grandma, a cousin, and lines between them. From such a drawing it is easy to see who is related to whom and how. A knowledge graph is the same kind of drawing, only huge: instead of relatives it has cities, books, people and things, and the lines show how they are linked.

In plain words

An ordinary table stores rows: client, phone, amount. A knowledge graph stores points and arrows between them. A point is an object: a person, a company, a product, a document. An arrow is a named link: works at, supplies, signed. Such triples add up to a network you can walk through. For example, you can ask which suppliers are linked to projects that missed deadlines, and the system follows the arrows from projects to contracts and on to companies. With tables, you would have to combine several reports by hand.

For business

This data model is useful when the value is in the links: who is whose contractor, which contracts depend on one supplier, which clients are connected through shared owners. A knowledge graph AI setup is used in counterparty checks, purchasing, search in technical documentation and in schemes like GraphRAG. The main costs are not in the database but in filling it: entities must be extracted from documents and systems, duplicates merged (Alpha LLC and LLC Alpha) and kept up to date. If sources contradict each other, the graph will confidently show a wrong link. Start with one question from management that has no quick answer now. Such questions are usually covered by an executive dashboard on data from the accounting system (1C in Russia) and the CRM.

For a specialist

A knowledge graph represents knowledge as a graph: nodes stand for entities, and edges are typed relations between them. In the RDF model (W3C RDF 1.1 specification, 2014), the basic unit is a triple of subject, predicate and object; queries to such graphs are written in SPARQL. An alternative is property graphs, with attributes on nodes and edges and languages like Cypher. An ontology sets the vocabulary of classes and relations. The survey by Hogan et al. (2020) covers data models and query languages, the role of schema, identity and context, deductive and inductive methods of deriving knowledge, and the creation, enrichment and quality assessment of graphs. A well-known industrial example is the Google Knowledge Graph, introduced on 16 May 2012 under the motto things, not strings, with more than 500 million objects and 3.5 billion facts. Together with LLMs, graphs are used for search (GraphRAG) and for fact-checking answers.

Example

The purchasing department builds a graph from contracts and extracts from the Russian state register of legal entities. To the question which suppliers belong to one owner, the graph answers in seconds, although these links are not visible in the tables.

Related terms

Questions and answers

How is a knowledge graph different from an ordinary database? A relational database stores tables, and the links between them have to be assembled with queries. A graph stores the links directly, so chains like company, owner, another company are found quickly.
Why does AI need a knowledge graph? A language model can invent a link that does not exist. A graph gives it checked facts and relations, and in the GraphRAG setup it helps answer questions about the whole document archive.

Sources

Where companies use it

Not sure whether your company needs this?

AI solutions map
Contact us