Similarity Search Is Not Relationship Traversal

Vector-based retrieval finds content that’s semantically similar to a query. Graph-based retrieval traverses explicit relationships between entities.

These solve different problems. Vector search excels at discovery when you don’t know exactly what you’re looking for. Graph traversal excels when you need to reason through connections, especially connections between things that aren’t semantically similar at all.

A drug and its side effects aren’t semantically similar. A regulation and the business process it affects aren’t semantically similar. But the relationships between them are exactly what matters for certain questions.

Multi-hop reasoning often requires connecting things that share no lexical overlap. Similarity search will miss these connections entirely.

Related: 07-molecule—vectors-vs-graphs, 05-atom—multi-hop-reasoning