dev.langchain4j.rag.content.aggregator

Class ReciprocalRankFuser

java.lang.Object

dev.langchain4j.rag.content.aggregator.ReciprocalRankFuser

public class ReciprocalRankFuser
extends Object

Implementation of Reciprocal Rank Fusion.
A comprehensive explanation can be found here.

Constructor Summary

Constructors

Constructor and Description
ReciprocalRankFuser()

Method Summary

Modifier and Type	Method and Description
static List<Content>	fuse(Collection<List<Content>> listsOfContents) Fuses multiple List<Content> into a single List<Content> using the Reciprocal Rank Fusion (RRF) algorithm with k=60.
static List<Content>	fuse(Collection<List<Content>> listsOfContents, int k) Fuses multiple List<Content> into a single List<Content> using the Reciprocal Rank Fusion (RRF) algorithm.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ReciprocalRankFuser

public ReciprocalRankFuser()

Method Detail

fuse

public static List<Content> fuse(Collection<List<Content>> listsOfContents)

Fuses multiple List<Content> into a single List<Content> using the Reciprocal Rank Fusion (RRF) algorithm with k=60.

Parameters:

listsOfContents - A Collection of List<Content> to be fused together.

Returns:

A single List<Content>, the result of the fusion.

fuse

public static List<Content> fuse(Collection<List<Content>> listsOfContents,
                                 int k)

Fuses multiple List<Content> into a single List<Content> using the Reciprocal Rank Fusion (RRF) algorithm.

Parameters:

listsOfContents - A Collection of List<Content> to be fused together.

k - A ranking constant used to control the influence of individual ranks from different ranked lists being combined. A common value used is 60, based on empirical studies. However, the optimal value may vary depending on the specific application and the characteristics of the data. A larger value diminishes the differences between the ranks, leading to a more uniform distribution of fusion scores. A smaller value amplifies the importance of the top-ranked items in each list. K must be greater than or equal to 1.

Returns:

A single List<Content>, the result of the fusion.