What is Knowledge Graph?

Most organisations keep their data in many places: a CRM, an ERP, a data lake, spreadsheets and a large pile of documents. A knowledge graph sits across those sources and links the records that describe the same customer, product or contract, so they can be queried together.

Once the data is connected, some things get much easier. Search can take context into account, recommendations can draw on both the product catalogue and customer behaviour, and an AI system can show where an answer came from by pointing to the path it followed through the graph.

Google, Amazon and the BBC all run large knowledge graphs behind their products. In financial services and pharma they are also used for lineage: showing where a figure came from and what it depends on, which is what auditors and regulators ask for.

A knowledge graph is made of statements called triples. Each one has a subject, a predicate and an object, for example Semantic Partners, provides, Ontology Engineering. Every entity and relationship has its own URI, so there is no doubt about which thing a statement refers to, and data from different sources can be linked.

An ontology sets out the types of entity (classes) and the types of relationship (properties) the graph uses. It also lets software reason over the data. If the ontology says every Subsidiary is a kind of Organisation, then anything true of organisations in general also applies to each subsidiary.

The data usually lives in a triplestore and is queried with SPARQL, which can follow relationships to any depth within a single query.

Financial services. A bank can map ownership chains across jurisdictions in a knowledge graph, so the know-your-customer and anti-money-laundering checks that used to mean weeks of manual research become queries.

Pharmaceutical R&D. A drug discovery team can link compounds, targets, pathways and trial results, which lets scientists check a hypothesis against everything the company already knows instead of assembling the evidence by hand.

Energy. An operator can connect well data, sensor readings and geological models in a graph built on the OSDU standard, so engineers see the subsurface data from different teams in one place.