Short definition

A representation of a body of knowledge as nodes connected by relations, usable by computation: traversal, grouping, path finding.

Full definition

In the tradition of the semantic web and search engines, a knowledge graph connects typed entities — people, places, products — through relations drawn from a controlled vocabulary, so as to allow inference.

In the articles on this blog, the term more modestly designates the graph you get from a set of notes or files: each note is a node, each link between two notes an edge, and the relation carries the free text that explains the link. There is neither controlled typing nor an inference engine; the value comes from what a tool can traverse, split and query.

Usage in the field

Used when tooling up a knowledge base of notes or a code repository — Obsidian, a wiki maintained by an AI, Graphify — at the point of choosing how to build the graph and what to expect from it.

Synonyms and variants

"Graph of notes", knowledge graph.

Not to be confused with

• Ontology — a description of a domain through its objects, their properties, their links and their rules, written to be reread and kept up to date. An ontology can be represented as a graph; a graph of notes does not describe a domain, it connects ideas.
• Obsidian's graph view — a display of the links between notes, with no computation of groups, no named relation, no query.
• vector database — an index where notes are brought together by proximity of meaning, with no written link between them.

Examples

A graph of 1,013 nodes and 6,442 links, built from a marketing wiki by a script, then explored with Graphify.

Ambiguities / debates

Semantic web practitioners reserve the term for typed graphs that can be queried with a dedicated language; common tool usage extends it to any graph built from documents, including by an AI that infers the relations.