Skip to main content
This section uses vector search, a beta feature available in versions 0.25.0 and above.
Hybrid search is a technique that combines the strengths of BM25 scoring and vector search to produce more relevant results. Because BM25 scores are produced by exact term matches while vector search matches on semantic meaning, a row missed by one method is often found by the other.

​
Fusing by Rank

Because BM25 scores and vector distances operate on different scales, the values cannot simply be added together. Instead, the standard approach is Reciprocal Rank Fusion, which discards the scores and uses only each row’s position, or rank, in its own result list. By using rank, a row that ranks first contributes the same amount whether its BM25 score was 1 or 100, so the two searches can be combined without ever being placed on a common scale.