Lesson 711 lessons
Hybrid Search — Semantic + Keyword
Where pure semantic search falls short
Semantic search can miss exact matches for specific codes, product SKUs, names, or acronyms — these benefit from traditional exact keyword matching that embeddings sometimes blur.
Combining both search types
Hybrid search runs both a keyword search (BM25 or similar) and a semantic vector search, then merges and re-ranks the combined results — capturing both exact-match precision and conceptual recall.
When hybrid search is worth the added complexity
For knowledge bases with a lot of technical terminology, codes, or exact-match-sensitive content (legal, medical, product catalogs), hybrid search is usually worth the extra implementation effort over pure semantic search.
Key Takeaways
- Pure semantic search can miss exact matches for codes, names, and acronyms.
- Hybrid search combines keyword matching with semantic vector search.
- This captures both exact-match precision and conceptual recall.
- Worth the added complexity for technical or exact-match-sensitive content.
Test a case where keyword search wins
Find a document containing a specific code or acronym, and test whether pure semantic search retrieves it reliably for a query using that exact term.