Reciprocal Rank Fusion
A parameter-free strategy for fusing retrieval results that takes the reciprocal of each document's rank in each result list plus a smoothing constant k, sums the values, and re-ranks the documents, avoiding the problem of incomparable scores across different retrieval paths.
Core Facts
RRF does not require normalization of raw scores from different retrievers and relies only on rank information, but when multiple retrievers systematically make the same type of error, RRF only amplifies the mistakes
The Reciprocal Rank Fusion (RRF) algorithm was proposed by Cormack et al. in 2009, with a core formula of Score = Σ 1/(k + rank_i), where k typically takes the value 60 as a smoothing constant.
Timeline (last 90 days)
Agentic GraphRAG finds paragraphs through Neo4j's native full-text index and vector index, and uses reciprocal rank fusion (RRF) to fuse keyword rankings and semantic rankings.
Hybrid search runs dense vector search and classic keyword search (BM25) simultaneously, and typically uses Reciprocal Rank Fusion to merge the results, with the fusion formula summing 1 / (k + rank) over each document's rank in each list.
Systems such as Elasticsearch and Weaviate all have RRF built in, and k=60 is the industry default.
The original Anthropic text mentions using 'rank fusion techniques' to merge the two result streams but does not name the specific method, and the script adopted reciprocal rank fusion with the constant k=60.
Reciprocal Rank Fusion (RRF) is insensitive to score scale; as long as a document consistently ranks high across multiple retrieval results, its final fusion score will improve, making it in practice more robust than manually tuning weights.
RRF does not require normalization of raw scores from different retrievers and relies only on rank information, but when multiple retrievers systematically make the same type of error, RRF only amplifies the mistakes
All Facts (9)
RRF does not require normalization of raw scores from different retrievers and relies only on rank information, but when multiple retrievers systematically make the same type of error, RRF only amplifies the mistakes
65%VerifiedThe Reciprocal Rank Fusion (RRF) algorithm was proposed by Cormack et al. in 2009, with a core formula of Score = Σ 1/(k + rank_i), where k typically takes the value 60 as a smoothing constant.
65%Unverifiednmem的混合搜索将向量语义搜索和传统关键词搜索结合，通常采用Reciprocal Rank Fusion或线性加权的融合策略
85%UnverifiedContext Mode使用Reciprocal Rank Fusion算法合并两种搜索策略的排序结果
85%UnverifiedAgentic GraphRAG finds paragraphs through Neo4j's native full-text index and vector index, and uses reciprocal rank fusion (RRF) to fuse keyword rankings and semantic rankings.
50%UnverifiedHybrid search runs dense vector search and classic keyword search (BM25) simultaneously, and typically uses Reciprocal Rank Fusion to merge the results, with the fusion formula summing 1 / (k + rank) over each document's rank in each list.
50%UnverifiedSystems such as Elasticsearch and Weaviate all have RRF built in, and k=60 is the industry default.
50%UnverifiedThe original Anthropic text mentions using 'rank fusion techniques' to merge the two result streams but does not name the specific method, and the script adopted reciprocal rank fusion with the constant k=60.
50%UnverifiedReciprocal Rank Fusion (RRF) is insensitive to score scale; as long as a document consistently ranks high across multiple retrieval results, its final fusion score will improve, making it in practice more robust than manually tuning weights.
50%