Hybrid search in AI Search

In hybrid search mode, AI Search blends keyword search and semantic vector search to find results that best match the terms and meaning of your search.

Overview of hybrid search

In older releases, AI Search performs all searches in keyword search mode. This mode looks for the best matches for your search terms, but doesn't take the context or meaning of those terms into account. The keyword search relevance score for a search result indicates how well the indexed terms in that search result match your search terms.

Starting in the Yokohama Patch 11 release, AI Search includes an alternate semantic vector search mode in features which work with the Now LLM Service. Semantic vector search analyzes the meanings and context of your search terms and uses that information to find results with similar meanings. It improves search recall by interpreting natural language to more accurately reflect your search's intent. The semantic vector search relevance score for a search result indicates how closely the search result matches the meaning of your search.

Note:

To learn more about semantic vector search mode and how it differs from keyword search mode, see Semantic vector search in AI Search.

Beginning with Now Assist in AI Search 15, installation of Now Assist in AI Search offers a new hybrid search mode. This mode combines keyword relevance scores and semantic similarity scores into a single result ranking using Reciprocal Rank Fusion (RRF). By blending keyword and semantic vector search modes, AI Search offers the best of both worlds, providing users with both search precision and contextual relevance.

Note:

Hybrid search applies to any indexed source that's included in the semantic index. By default, the following indexed sources ship with preconfigured semantic indexing and work with hybrid search:

• Knowledge Table indexed source [kb_knowledge table]
• Catalog Item Table indexed source [sc_cat_item table]
• Skills indexed source [sys_gen_ai_skill table]
• All External Content Connectors indexed sources [connector-specific tables]

You can manually configure additional indexed sources to be included in the semantic index, extending hybrid search to those sources as well.

Benefits of hybrid search

Hybrid search offers these benefits when compared with keyword search.

Improved search quality and recall

Hybrid search combines keyword and semantic matching, increasing the chances of retrieving relevant results, even when users phrase their queries differently.

More relevant top-ranked search results

Semantic understanding helps surface answers that better match the user’s intent instead of just their search keywords.

Fewer zero-result searches

Hybrid search can return useful results even for vague or misspelled queries because it uses semantic matching to understand meaning beyond exact keywords. This reduces the number of searches that return no results.

Availability of hybrid search

Hybrid search requires installation of the Now Assist in AI Search application. This feature is available beginning in Now Assist in AI Search 15.

If you install Now Assist in AI Search without having any previous version installed, hybrid search is automatically activated for your search application configurations that use AI Search as their search engine.

Note:

Search application configurations that use the legacy Zing text indexing and search engine don't support hybrid search.

If you upgrade to Now Assist in AI Search from a previous version, hybrid search isn't automatically activated for your search application configurations that use AI Search as their search engine. To learn how to manually activate hybrid search for these search application configurations in the AI Search Admin console, see Manage hybrid search in search applications.

Interactions with other features

Activating hybrid search in a search application configuration disables search result counts for facets in that search application configuration. For more information on search result counts for facets, see Show search result counts for facets on the results page for a search application.

When you activate hybrid search in a search application configuration, the behavior of the Most recent search result sort option changes for the specified search application. Instead of sorting the entire list of search results by last modification date, AI Search first retrieves keyword search results and the most relevant semantic vector search results, then merges those results and sorts them by last modification date. To learn about sorting search results based on recency, see Change the sort order for your search results.

• Activate hybrid search for a search application configuration in AI Search
  Activate the hybrid search mode for a search application configuration that uses AI Search as its search engine. Hybrid search offers improved precision and recall for knowledge article, Catalog Item, external content, and topic retrieval searches.