On the Embedding Collapse when Scaling up Recommendation Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Guo, Xingzhuo, Pan, Junwei, Wang, Ximei, Chen, Baixu, Jiang, Jie, Long, Mingsheng
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914826110894080
author Guo, Xingzhuo
Pan, Junwei
Wang, Ximei
Chen, Baixu
Jiang, Jie
Long, Mingsheng
author_facet Guo, Xingzhuo
Pan, Junwei
Wang, Ximei
Chen, Baixu
Jiang, Jie
Long, Mingsheng
contents Recent advances in foundation models have led to a promising trend of developing large recommendation models to leverage vast amounts of available data. Still, mainstream models remain embarrassingly small in size and naïve enlarging does not lead to sufficient performance gain, suggesting a deficiency in the model scalability. In this paper, we identify the embedding collapse phenomenon as the inhibition of scalability, wherein the embedding matrix tends to occupy a low-dimensional subspace. Through empirical and theoretical analysis, we demonstrate a \emph{two-sided effect} of feature interaction specific to recommendation models. On the one hand, interacting with collapsed embeddings restricts embedding learning and exacerbates the collapse issue. On the other hand, interaction is crucial in mitigating the fitting of spurious features as a scalability guarantee. Based on our analysis, we propose a simple yet effective multi-embedding design incorporating embedding-set-specific interaction modules to learn embedding sets with large diversity and thus reduce collapse. Extensive experiments demonstrate that this proposed design provides consistent scalability and effective collapse mitigation for various recommendation models. Code is available at this repository: https://github.com/thuml/Multi-Embedding.
format Preprint
id arxiv_https___arxiv_org_abs_2310_04400
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle On the Embedding Collapse when Scaling up Recommendation Models
Guo, Xingzhuo
Pan, Junwei
Wang, Ximei
Chen, Baixu
Jiang, Jie
Long, Mingsheng
Machine Learning
Information Retrieval
Recent advances in foundation models have led to a promising trend of developing large recommendation models to leverage vast amounts of available data. Still, mainstream models remain embarrassingly small in size and naïve enlarging does not lead to sufficient performance gain, suggesting a deficiency in the model scalability. In this paper, we identify the embedding collapse phenomenon as the inhibition of scalability, wherein the embedding matrix tends to occupy a low-dimensional subspace. Through empirical and theoretical analysis, we demonstrate a \emph{two-sided effect} of feature interaction specific to recommendation models. On the one hand, interacting with collapsed embeddings restricts embedding learning and exacerbates the collapse issue. On the other hand, interaction is crucial in mitigating the fitting of spurious features as a scalability guarantee. Based on our analysis, we propose a simple yet effective multi-embedding design incorporating embedding-set-specific interaction modules to learn embedding sets with large diversity and thus reduce collapse. Extensive experiments demonstrate that this proposed design provides consistent scalability and effective collapse mitigation for various recommendation models. Code is available at this repository: https://github.com/thuml/Multi-Embedding.
title On the Embedding Collapse when Scaling up Recommendation Models
topic Machine Learning
Information Retrieval
url https://arxiv.org/abs/2310.04400