Knowledge Graph Search
Search infrastructure that represents entities and their relationships as a graph, enabling structured navigation and semantic retrieval beyond keyword matching.
Core Idea
A knowledge graph stores:
- Entities — people, places, things (nodes)
- Relationships — typed edges connecting entities
- Properties — attributes of entities
Search over a knowledge graph enables:
- Entity disambiguation (“Apple” the company vs. the fruit)
- Relationship traversal (“movies with actors who worked with Spielberg”)
- Faceted filtering from structured data
Making a Graph Searchable
Netflix’s approach (federated graph):
- Graph construction — ingest structured metadata into entity nodes
- Embedding entities — represent nodes as vectors for similarity search
- Federated queries — route sub-queries to appropriate graph partitions
- Materialized views — pre-compute common traversals for latency
Graph + Vector Integration
Modern systems combine:
- Vector similarity for semantic retrieval
- Graph traversal for relational filtering
- The intersection: embed graph neighborhoods, not just leaf documents
Industry Examples
- Netflix: federated content graph searchable via vector embeddings
- LinkedIn: entity graph for people/job/company search
- Google Knowledge Panel: structured entity data surfacing