Package dev.langchain4j.store.embedding

Class CosineSimilarity

java.lang.Object

dev.langchain4j.store.embedding.CosineSimilarity

public class CosineSimilarity
extends Object

Utility class for calculating cosine similarity between two vectors.

Field Summary

Fields

Modifier and Type	Field	Description
static final float	EPSILON	A small value to avoid division by zero.

Method Summary

Modifier and Type	Method	Description
static double	between(Embedding embeddingA, Embedding embeddingB)	Calculates cosine similarity between two vectors.
static double	fromRelevanceScore(double relevanceScore)	Converts relevance score into cosine similarity.

Methods inherited from class java.lang.Object

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

Field Details

EPSILON

public static final float EPSILON

A small value to avoid division by zero.

See Also:

• Constant Field Values

Method Details

between

public static double between(Embedding embeddingA,
 Embedding embeddingB)

Calculates cosine similarity between two vectors.

Cosine similarity measures the cosine of the angle between two vectors, indicating their directional similarity. It produces a value in the range:

-1 indicates vectors are diametrically opposed (opposite directions).

0 indicates vectors are orthogonal (no directional similarity).

1 indicates vectors are pointing in the same direction (but not necessarily of the same magnitude).

Not to be confused with cosine distance ([0..2]), which quantifies how different two vectors are.

Embeddings of all-zeros vectors are considered orthogonal to all other vectors; including other all-zeros vectors.

Parameters:

embeddingA - first embedding vector

embeddingB - second embedding vector

Returns:

cosine similarity in the range [-1..1]

fromRelevanceScore

public static double fromRelevanceScore(double relevanceScore)

Converts relevance score into cosine similarity.

Parameters:

relevanceScore - Relevance score in the range [0..1] where 0 is not relevant and 1 is relevant.

Returns:

Cosine similarity in the range [-1..1] where -1 is not relevant and 1 is relevant.