Rank Fusion Settings

Overview

The Rank Fusion feature of Fess integrates multiple search results to provide more accurate search results.

What is Rank Fusion

Rank Fusion is a technique that combines results from multiple search algorithms or scoring methods to generate a single optimized ranking.

Key benefits:

Combines the strengths of different algorithms
Improves search accuracy
Provides diverse search results

Supported Algorithm

Fess uses the RRF (Reciprocal Rank Fusion) algorithm for Rank Fusion.

RRF (Reciprocal Rank Fusion)

RRF calculates scores by summing the reciprocal of each result’s rank.

Formula:

score(d) = Σ 1 / (k + rank(d))

k: Rank constant parameter (default: 20)
rank(d): Rank of document d in each search result

Settings

fess_config.properties

Configuration properties:

# Window size for rank fusion
rank.fusion.window_size=200

# Rank constant (k parameter for RRF)
rank.fusion.rank_constant=20

# Number of threads (-1 for auto)
rank.fusion.threads=-1

# Score field name in results
rank.fusion.score_field=rf_score

Property	Default	Description
`rank.fusion.window_size`	`200`	Number of results to consider from each search type for fusion
`rank.fusion.rank_constant`	`20`	The k constant in the RRF formula; controls how much weight is given to lower-ranked results
`rank.fusion.threads`	`-1`	Number of threads for parallel execution (-1 for automatic)
`rank.fusion.score_field`	`rf_score`	Field name used to store the computed rank fusion score

Usage Examples

Basic Hybrid Search

Calculate BM25 score with keyword search
Calculate vector similarity with semantic search
Fuse both results with RRF
Generate final ranking

Search Flow:

User Query
    ↓
┌──────────────────┬──────────────────┐
│  Keyword Search  │ Semantic Search  │
│    (BM25)        │  (Vector)        │
└────────┬─────────┴────────┬─────────┘
         ↓                  ↓
     Rank List 1        Rank List 2
         └────────┬─────────┘
                  ↓
          Rank Fusion (RRF)
                  ↓
          Final Ranking

Performance Considerations

Adjust rank.fusion.window_size to balance quality and speed; larger values consider more candidates but use more memory
Set rank.fusion.threads to control parallelism; -1 automatically determines the thread count
Memory usage increases with larger window sizes as more search results are retained

Troubleshooting

Search Results Differ from Expectations

Symptom: Results after Rank Fusion differ from expectations

Checks:

Verify results from each search type individually
Adjust the rank.fusion.rank_constant value
Adjust rank.fusion.window_size to include more candidate results

Out of Memory

Symptom: OutOfMemoryError occurs

Solutions:

Reduce rank.fusion.window_size
Increase JVM heap size
Disable unnecessary search types

Reference

Scripting Overview - Scripting Overview
../admin/search-settings - Search Settings Guide
LLM Integration Overview - LLM Integration Guide (Semantic Search)