# Hybrid Search

# Hybrid Search

Hybrid search combines text search and vector search to get the best of both worlds:

• precision from keyword matching
• recall from semantic similarity

This is often the best default for enterprise datasets that mix identifiers, structured terms, and long unstructured content.

# Common hybrid patterns

• Keyword-first, semantic re-rank

  • run text search to get candidates
  • re-rank candidates using embedding similarity

• Keyword + semantic candidate union

  • retrieve candidates from text and vector search
  • merge and deduplicate results
  • apply ranking rules and facets

• Context-constrained semantic search

  • use graph traversal or facets to define a target set
  • run semantic similarity only within that set

# Why graph context matters

Hybrid search becomes significantly more useful when combined with graph constraints:

• “search within this customer”
• “search within this product line”
• “search within tickets related to this device”

Graph constraints reduce noise and make semantic retrieval safer.

# Evaluation guidance

To tune hybrid search, evaluate with real queries:

• top queries by volume
• “zero result” queries from keyword-only search
• queries with strong domain vocabulary (acronyms, product names)

Measure:

• precision@k (are the first results good?)
• recall (did we retrieve the right items at all?)
• user refinement behavior (do facets help?)

# Next steps

• Learn the tuning levers: Ranking Tuning
• Configure the underlying engines: Text Search and Vector Search