D
Doc and Tell
Glossary/ai/ml
ai/ml

Reciprocal Rank Fusion (RRF)

An algorithm that combines ranked result lists from multiple retrieval methods into a single merged ranking without requiring score normalization.

RRF assigns each document a score of 1/(k + rank) for each retrieval method, where k is a smoothing constant (typically 60) and rank is the document's position in that method's result list. The per-method scores are summed across all methods, and documents are sorted by combined score. RRF is particularly valuable in hybrid retrieval because vector search scores and BM25 scores exist on incompatible scales — normalizing them requires calibration that may not generalize across query types.

RRF consistently outperforms score-based fusion in benchmark evaluations and is robust to the specific choice of retrieval methods being combined. In document intelligence systems, RRF is typically applied to merge vector search results (high semantic relevance) and BM25 results (high keyword precision), ensuring that documents ranking highly in either method are promoted in the final result set. This hybrid approach captures both the semantic understanding of dense retrieval and the exact-match reliability of sparse retrieval.

Analyze Documents Related to Reciprocal Rank Fusion (RRF)

Upload any document and get AI-powered analysis with verifiable citations.

Start Free