GLOSSARY

Knowledge Graph

A knowledge graph connects entities (people, products, brands) with relationships. In e-commerce, it links products to attributes, guides, and compatibility so search and recommendations make sense.

What is a Knowledge Graph?

A knowledge graph (KG) is a structured network of entities (nodes) and relationships (edges) described with IDs, types, and properties. It encodes facts like Brand → makes → Product, Product → hasMaterial → Merino, or Shoe → compatibleWith → Insole that systems can use for retrieval, QA, and reasoning.

How It Works (quick)

  • Model: Define schema/ontology (classes, relations, constraints); align with taxonomy and controlled vocabulary.
  • Ingest & link: Extract entities from text/feeds; entity-link to canonical IDs; resolve duplicates; store provenance.
  • Store & query: Use graph DB/triple store; query with SPARQL/GraphQL or APIs.
  • Use in search: Expand queries via relations (brand↔collection), enrich snippets, enforce compatibility, and power Best Bets.
  • Maintain: Version entities/relations, track confidence, and run integrity checks.

Why It Matters in E-commerce

  • Disambiguation: Distinguish “Apple” the brand from the fruit; “GTX” as GORE-TEX.
  • Better discovery: Generate collections from graph patterns (e.g., waterproof trail shoes).
  • Compatibility & bundles: Ensure fits-with recommendations (printer ↔ cartridge).
  • Unified answers: Connect PDPs with guides, policies, and FAQs for richer SERPs and onsite answers.
  • Analytics: Clean, joinable dimensions (brand/material/use-case) for reporting.

Best Practices

  • Canonical IDs: Stable IDs per entity; never key on labels alone.
  • Sources & provenance: Keep where facts came from and when; add confidence scores.
  • Governance: Editorial review for sensitive relations; deprecation maps; release notes.
  • Interoperability: Reuse public schemas (Schema.org, GS1) and align with your taxonomy.
  • Freshness: Event-driven updates when products, attributes, or relationships change.
  • Privacy & ACLs: Attach permissions where needed (e.g., B2B assortments).

Challenges

  • Data quality: Wrong merges, near-duplicates, stale relations.
  • Scale & cost: Extract/link at volume; graph queries under low latency.
  • Change management: Ontology drift; cross-team adoption.

Examples

  • Link Shoe → hasMaterial → GORE-TEX; Shoe → terrain → trail → auto-create a Trail Waterproof Shoes collection.
  • Map Help Article → answers → returns policy and surface it with related PDPs.
  • Connect Laptop → compatibleWith → RAM module for accurate accessories.

Summary

A knowledge graph turns scattered facts into a connected model your search and recommendations can trust—improving disambiguation, compatibility, discovery, and answer quality.

FAQ

Knowledge graph vs taxonomy?

Taxonomy is a hierarchy; a KG is a graph that models many relation types (not just parent/child).

Knowledge graph vs vector DB?

Vectors capture similarity; KGs capture explicit facts/relations. Many stacks use both.

Do I need a graph database?

Not always; you can store edges in relational form, but native graphs simplify traversal-heavy use cases.