Retrieval

What is Hybrid Search?

Hybrid search combines keyword-based search (like BM25) with semantic vector search, then fuses results using techniques like Reciprocal Rank Fusion. This catches both exact matches and conceptually similar content—often outperforming either approach alone.

Why It Matters

Semantic search alone misses exact terms like product codes ("SKU-12345") or legal citations ("Section 4.2.1"). Keyword search misses synonyms and paraphrases. Hybrid search gets the best of both—critical for enterprise RAG where precision matters.

How Hybrid Search Works

1
Keyword Search — BM25 finds exact term matches, returns ranked list
2
Vector Search — Embedding similarity finds semantic matches
3
Fusion — Combine rankings (RRF, weighted sum, or learned)

When Each Approach Wins

Query TypeBest ApproachExample
Exact identifiersKeyword"Error code E-4012"
Conceptual questionsSemantic"How to reduce costs"
Mixed queriesHybrid"GDPR Article 17 data deletion"

Related Terms

← Back to Glossary