Why it matters
Semantic search matches meaning when users do not share your vocabulary. Customers say money back while policy says pro-rata credit. Keyword-only search misses both.
How it works
Embed queries and documents into vectors, find nearest neighbors, combine with metadata filters. Tune embedding models on domain language where generic models underperform.
Example
Nintendo semantic search maps cancel my subscription to Nintendo Help Center cancellation and Refund Policy sections even when users never type refund.
Common mistakes
- 1Semantic-only search missing exact SKU and error codes
- 2Stale embeddings after major product rename
- 3No evaluation for cross-language queries in global products
Definitions are useful. Governed knowledge is better.
Wiki helps you turn these concepts into a real system for the knowledge behind your AI.
See Wiki in action