What does Knowledge Graph mean?

A knowledge graph stores facts as a network of nodes and edges. Nodes stand for individual things such as people, places, products, or cases; edges stand for the relationship between them. Unlike a table with fixed columns, a knowledge graph can be extended with new kinds of relationships without touching the existing entries.

Every statement is stored as a triple of two things and the relationship between them — for instance, that contract 4711 was signed with the company Meier. An ontology defines which types of nodes and relationships are allowed. Queries run through graph databases such as Neo4j or through SPARQL, and even multi-step connections can be searched. That is precisely where the difference to a relational database lies: there, every additional step requires another join in the query.

A knowledge graph pays off where the relationships between entries matter more than the entries themselves. Examples include supply chains, ownership structures, areas of responsibility, and linking files, people, and deadlines. For a closed dataset of uniform records, a table remains the simpler solution.

The advantage over distributed individual systems lies in multi-step queries. A question about every case connected two steps removed to a specific company is a single line in a graph. New data sources can be attached by connecting their entries to existing nodes.

Every connection in the graph needs to be captured when an entry is created. If it’s forgotten, the node won’t turn up in any query, even though the entry itself exists. That’s why a regular check scans the dataset for nodes without connections and for contradictory entries.