Skip to main content
Version: 2.0

Hybrid search

Hybrid search blends semantic matching (meaning) and traditional keyword matching (exact terms) give you more control over retrieval. It helps you handle cases where meaning is important, but specific terms still matter.

You control this balance using the lexical_interpolation value.

How semantic matching works​

Semantic matching retrieves content based on meaning instead of exact words. Vectara represents both your uploaded documents and your queries as embeddings. These embeddings (vectors) capture the semantic intent that can identify results even with different wording.

  1. Indexing: You upload documents, Vectara breaks them into smaller chunks.
    This generates an embedding for each chunk, where chunks with similar meaning are grouped nearby in the embedding space. For example, the phrases "fix a flat tire" and "change a wheel" use different words but get grouped closely.
  2. Querying: A user submits a query, and Vectara generates an embedding for that query, which captures the intent behind the query.
  3. Matching: Vectara compares the query embedding to the indexed chunks and retrieves the ones closest in meaning. Because this comparison is semantic, the search results can match intent even when wording differs.

Vectara uses semantic search by default.

How hybrid search works​

Hybrid search combines semantics with keywords to produce a single result. The semantic score measures how closely the chunk's meaning matches the query, while the lexical score measures how closely the keywords align with the query.

Hybrid search interpolates these signals to give you the best of both worlds:

  • When specific keywords must carry weight such as product names, SKUs, and IDs.
  • When your domain includes schema names, commands, or jargon.
  • When you need meaningful results but certain keywords should boost ranking.

If you're not sure where to start, begin with semantic search and modify the lexical_interpolation value until the results match your expectations.

You enable hybrid search by specifying the lexical_interpolation value in the search object of a query:

  • 0.0: Pure semantic search.
  • 1.0: Pure keyword search.
  • 0.0 to 1.0: A hybrid search that balances both methods.

A common starting point is 0.025.

note

A lexical_interpolation value of 1.0 is equivalent to traditional BM25 keyword search.

Example:

LEXICAL INTERPOLATION EXAMPLE

Code example with json syntax.
1
tip

💡 You can test queries with different lexical_interpolation values in our API Reference and in the Vectara Console.

Syntax interpretation​

When interpreting query strings, Vectara treats the following syntax specially.

  • Words that are quoted must match exactly in that order. For example, the query blue shoes must match the word blue followed immediately by shoes.

  • A word fragment suffixed with an asterisk * is treated as a prefix match, meaning that it matches any word of which it is a prefix. For example, Miss* matches Mississippi.

  • Words prefixed with a minus - sign are excluded from the results. To extend on the previous example, -Mississippi would exclude results referencing the Magnolia State. Using -Miss* would exclude references to both Mississippi and Missouri.