Graph data and knowledge graphs

Some questions are about relationships, who is connected to whom, and graphs answer them better than tables. This guide covers graph data and knowledge graphs built from external data.

When relationships are the point

For fraud rings, ownership networks, supply chains and recommendations, the connections matter more than individual records. Graph data models entities and the relationships between them, making these patterns visible.

What a knowledge graph is

A knowledge graph organises entities and relationships into a connected, queryable structure, often integrating many sources into a single semantic model.

Building one from external data

Building a graph means resolving entities across sources (so the same company is one node), modelling relationships, and keeping it current. Entity resolution is central, errors here corrupt the whole graph.

Standards

Graph standards such as RDF and property-graph models, with query languages like SPARQL and Cypher, support interoperability and analysis.

Common use cases

Fraud and AML network analysis, beneficial-ownership and KYC, supply-chain mapping, and recommendation and intelligence.

Sourcing and governance

Graphs combine many sources, so provenance per edge and node matters, and personal data brings the GDPR into scope. A managed partner can source, resolve and assemble graph-ready data.

When a graph beats a table

Graphs earn their keep when the questions are about relationships and paths, not single records: who ultimately owns this company, which accounts connect to a flagged one, how a disruption propagates through a supply network. These multi-hop questions are awkward and slow in tables but natural in a graph. If your questions are mostly aggregations over rows, a table is simpler; if they traverse connections, a graph is the right model.

Building a graph from external data

The hard part is not the graph database but the data: resolving entities across sources so the same company is one node, modelling relationships consistently, and keeping provenance per node and edge. Entity-resolution errors corrupt the whole graph, a wrong merge creates false connections that look like insight. Standards such as RDF and property-graph models, with query languages like SPARQL and Cypher, then make the result interoperable and queryable.

Sources & further reading

• W3C: RDF and SPARQL standards.
• Industry references on property graphs and knowledge graphs.
• DAMA-DMBOK: data modelling and integration.
• EUR-Lex: Regulation (EU) 2016/679 (GDPR).