What is Knowledge graph?

There is no single formal specification for knowledge graphs, and the cited page is one vendor's description. It comes from Neo4j and says that definitions vary by audience. The page notes that triple stores based on the Resource Description Framework (RDF) use a different, more granular design from property graphs.

The Neo4j page describes entities such as people, places, products or organizations as nodes. Each node has labels showing its type and can carry properties. Relationships link two nodes, have a type, and can also carry properties. A customer connected to an order by a "placed order" relationship is the page's example. Beyond the data itself, a knowledge graph holds organizing principles, which are flexible rules or categories giving the data a conceptual structure. Ontologies, formal specifications of concepts in a subject area, are one kind of organizing principle, and the page says they can be costly to define and maintain.

The page lists uses including search, fraud detection, master data management, supply chains and drug discovery. For language models, it says a knowledge graph can ground answers through GraphRAG, which it claims improves accuracy and explainability.

A caveat from the page: knowledge graphs need not be large or elaborate, and it suggests starting with a small, use-case-specific scope. Its comparisons favoring property graphs over RDF triple stores reflect the vendor's own perspective.