Hybrid search
Hybrid search is a retrieval method that combines keyword matching, such as BM25, with vector similarity between embeddings and merges both result lists into one ranking. Keyword matching finds exact names, codes and rare words, and vector search finds pages that express the same idea in other words.
Many vector databases and enterprise search tools offer hybrid search, often as the default. A common way to merge the two lists is reciprocal rank fusion (RRF), introduced by Cormack, Clarke and Büttcher at SIGIR 2009, which scores each document by summing 1/(k + rank) over the lists it appears in.
Example
For the query “error E1234 on a dishwasher”, BM25 finds manuals that contain the exact code E1234, while vector search finds forum posts about a dishwasher that will not drain. Hybrid search can rank the manual first and the most useful forum thread second.
FAQ
Is hybrid search better than vector search?
Hybrid search usually does better than vector search alone on queries with exact names, codes or rare words, because the keyword part matches them directly. On paraphrased questions the gain is smaller, since vector search already handles them.
What is reciprocal rank fusion?
Reciprocal rank fusion is a way to merge ranked lists in hybrid search: each document gets the sum of 1/(k + rank) across the lists it appears in, with k often set to 60, and the merged list is sorted by that sum.
Related terms
Written by Keenable.
Published
