Is Cosine-Similarity of Embeddings Really About Similarity?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Steck, Harald, Ekanadham, Chaitanya, Kallus, Nathan
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917608390918144
author Steck, Harald
Ekanadham, Chaitanya
Kallus, Nathan
author_facet Steck, Harald
Ekanadham, Chaitanya
Kallus, Nathan
contents Cosine-similarity is the cosine of the angle between two vectors, or equivalently the dot product between their normalizations. A popular application is to quantify semantic similarity between high-dimensional objects by applying cosine-similarity to a learned low-dimensional feature embedding. This can work better but sometimes also worse than the unnormalized dot-product between embedded vectors in practice. To gain insight into this empirical observation, we study embeddings derived from regularized linear models, where closed-form solutions facilitate analytical insights. We derive analytically how cosine-similarity can yield arbitrary and therefore meaningless `similarities.' For some linear models the similarities are not even unique, while for others they are implicitly controlled by the regularization. We discuss implications beyond linear models: a combination of different regularizations are employed when learning deep models; these have implicit and unintended effects when taking cosine-similarities of the resulting embeddings, rendering results opaque and possibly arbitrary. Based on these insights, we caution against blindly using cosine-similarity and outline alternatives.
format Preprint
id arxiv_https___arxiv_org_abs_2403_05440
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Is Cosine-Similarity of Embeddings Really About Similarity?
Steck, Harald
Ekanadham, Chaitanya
Kallus, Nathan
Information Retrieval
Machine Learning
Cosine-similarity is the cosine of the angle between two vectors, or equivalently the dot product between their normalizations. A popular application is to quantify semantic similarity between high-dimensional objects by applying cosine-similarity to a learned low-dimensional feature embedding. This can work better but sometimes also worse than the unnormalized dot-product between embedded vectors in practice. To gain insight into this empirical observation, we study embeddings derived from regularized linear models, where closed-form solutions facilitate analytical insights. We derive analytically how cosine-similarity can yield arbitrary and therefore meaningless `similarities.' For some linear models the similarities are not even unique, while for others they are implicitly controlled by the regularization. We discuss implications beyond linear models: a combination of different regularizations are employed when learning deep models; these have implicit and unintended effects when taking cosine-similarities of the resulting embeddings, rendering results opaque and possibly arbitrary. Based on these insights, we caution against blindly using cosine-similarity and outline alternatives.
title Is Cosine-Similarity of Embeddings Really About Similarity?
topic Information Retrieval
Machine Learning
url https://arxiv.org/abs/2403.05440