RRF (Reciprocal Rank Fusion)

In depth

Reciprocal Rank Fusion is a search-engineering technique for merging the result lists from two or more independent retrieval methods. Each method assigns a rank to each candidate result; RRF turns those ranks into a single combined score using a formula that weights higher-ranked results more strongly while preventing any one method from dominating. In e-commerce search, RRF is the standard way to combine BM25 (keyword) and vector (semantic) results so a query like 'cheap warm winter jacket' returns products that match both the keyword 'jacket' and the underlying meaning 'insulated outerwear'. Whether fusion beats a single retrieval method depends on the catalog, so it is worth measuring rather than assuming.