Reciprocal Rank Fusion (RRF)

Gold definitionUpdated Apr 2, 2026

Definition

Reciprocal Rank Fusion (RRF) is a rank aggregation technique that combines ranked lists from multiple retrieval systems by summing the reciprocal of each document's rank, effectively merging diverse evidence to improve overall retrieval performance.

At a glance

Executive summary

Reciprocal Rank Fusion (RRF) is a method to combine search results from different systems into one improved list. It works by giving higher scores to items that appear high in any of the individual search results, making the final list more comprehensive and reliable. This is especially useful for complex questions where multiple search strategies are needed.

TL;DR

Reciprocal Rank Fusion combines ranked lists from different search engines or retrieval methods to create a better, more comprehensive single list.

Key points

Aggregates ranked lists by summing reciprocal ranks (1/rank) from multiple retrieval systems.
Improves retrieval performance and evidence coverage by combining diverse search signals, especially for complex queries.
Used by information retrieval researchers, developers of hybrid search systems, and RAG frameworks in domains like medical QA.
Unlike simple union or intersection, RRF intelligently weights items based on their rank across systems, being more robust than methods requiring score normalization.
Increasingly used in Retrieval-Augmented Generation (RAG) to combine sparse and dense retrievers for more robust and comprehensive context.

Use cases

Medical Question Answering: Combining BM25 (sparse) and Contriever (dense) results to provide comprehensive evidence for LLMs in clinical scenarios.
Enterprise Search: Merging results from keyword search, semantic search, and knowledge graph queries to provide a unified, highly relevant search experience for internal documents.
Meta-Search Engines: Aggregating results from multiple web search engines (e.g., Google, Bing, DuckDuckGo) to present a single, diverse result page.
E-commerce Product Search: Combining results from text-based product descriptions, image similarity search, and user review analysis to rank products more effectively.
Legal Document Discovery: Fusing ranked lists from different legal databases or search tools to ensure all relevant precedents and statutes are identified.

Also known as

RRF, Rank Fusion