Reciprocal Rank Fusion
A rank-aggregation method that combines multiple result lists by summing reciprocal functions of item rank rather than raw scores.
Information-Retrieval Context
Reciprocal rank fusion combines several ranked result lists by accumulating a reciprocal function of each item's rank. Because it operates on ranks rather than raw model scores, it is useful when lexical and dense retrievers produce values on incompatible scales.
Ranking Boundary
RRF is an aggregation rule, not a content-aware reranker. It does not re-encode a query-document pair or generate a new semantic relevance score.