Reciprocal Rank Fusion Explainer
Reciprocal rank fusion, or RRF, combines result lists from multiple retrieval systems. Each item receives points based on the reciprocal of a constant plus its position in each list, and the points are summed.
Because RRF uses ranks rather than raw scores, it avoids having to calibrate lexical, vector, or specialized retrieval scores onto one numerical scale. Items ranked highly by several systems tend to rise.
RRF is a practical baseline for hybrid search, ensemble retrieval, and federated search. It is simple, does not require labeled training data, and can merge lists of different lengths.
The rank constant controls how strongly the top positions dominate. RRF ignores the size of score gaps and may reward correlated systems repeatedly. Teams should compare constants, deduplicate items, and evaluate fusion by query type and business outcome.
Reciprocal rank fusion combines several result lists by rewarding items that appear near the top of one or more rankings.
Key terms include reciprocal rank fusion, RRF, rank, result list, fusion score, rank constant, ensemble retrieval and deduplication.
Every system contributes more for items near the top of its list.
Summed reciprocal ranks create one ordering without score normalization.
The constant changes the influence of very high versus moderately high positions.
Highly similar retrieval systems may add redundant evidence rather than independent support.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.
