Reciprocal Rank Fusion (RRF)
Algorithm that combines results from multiple retrieval methods into a single ranked list using inverse rank scores.
Last updated: April 26, 2026
Definition
Reciprocal Rank Fusion (Cormack et al. 2009) is the standard method for combining results from multiple retrievers. For each document, sum 1 / (rank + k) across each retriever's result list (k is a constant, typically 60). Documents that appear high in multiple lists get the highest combined scores. Documents that appear in only one list still rank, just lower. RRF is the workhorse of hybrid search because it requires no normalization between disparate scoring scales (dense similarity scores and BM25 scores are not comparable; ranks are).
When To Use
Use RRF whenever combining retrieval results from multiple sources. It is the default for hybrid search and for combining cross-encoder rerank with bi-encoder retrieval.
Related Terms
Building with Reciprocal Rank Fusion (RRF)?
I've shipped this pattern in real production systems. If you want a second pair of eyes on your architecture, that's what I do.