Reciprocal Rank Fusion (RRF)

Overview

Reciprocal Rank Fusion (RRF) is an algorithm that evaluates search scores from multiple, previously ranked results to produce a unified result set, merging and homogenizing the rankings into a single result set for the query response.

Key Advantages

Normalization-Free

You don't need to know the distribution of your vector or BM25 scores as RRF works purely on position, avoiding score distribution issues by focusing exclusively on rank positions.

Outlier Resistance

Because RRF aggregates rankings rather than scores, it prevents anomalous values from distorting relevance.

Scalability

Extremely efficient for sharded, billion-scale indices where global score normalization is expensive.

How It Works

For each document in search results, the engine assigns a reciprocal rank score based on its position:

Formula: score = 1/(rank + k)

Where:

• rank = position of the document
• k = constant (experiments show best results with k=60)

2026 Implementations

OpenSearch 2.19

Introduces RRF as a new feature in the Neural Search plugin that enhances hybrid search by merging ranked results from multiple query sources (neural search, k-NN, Boolean queries).

Azure AI Search

RRF is used when two or more queries execute in parallel, for hybrid queries and multiple vector queries.

Google Cloud Vertex AI

Uses RRF to merge token-based and semantic search results.

Common Use Cases

RRF is an excellent "first stage" reranker. Common pattern:

1. Retrieve top 100 documents via RRF
2. Use more expensive Cross-Encoder to rank top 10 for LLM context window

Best Practices

RRF is the best starting point for hybrid search because of its simplicity and resilience to mismatched score scales.