On the Mechanisms of Collaborative Learning in VAE Recommenders

2025-11-10

AI Summary

This work addresses three key challenges in Variational Autoencoder (VAE)-based collaborative filtering: excessive local collaboration, insufficient global collaboration, and posterior collapse in user representations. We propose anchor regularization—a novel framework that explicitly models latent-space proximity and semantic relevance among users by aligning user posterior distributions with item embeddings, augmented by β-KL regularization, input masking, and Wasserstein distance metrics. This design enhances cross-item information sharing among distant users while preserving individual user identity. Theoretical analysis characterizes how latent-space geometry governs collaborative learning dynamics. Extensive experiments on Netflix, MovieLens-20M, and Million Song datasets demonstrate consistent improvements in recommendation accuracy. Furthermore, the algorithm has been deployed in production on Amazon’s streaming platform, where online A/B testing confirms statistically significant gains in key recommendation metrics.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Variational Autoencoders (VAEs) are a powerful alternative to matrix factorization for recommendation. A common technique in VAE-based collaborative filtering (CF) consists in applying binary input masking to user interaction vectors, which improves performance but remains underexplored theoretically. In this work, we analyze how collaboration arises in VAE-based CF and show it is governed by latent proximity: we derive a latent sharing radius that informs when an SGD update on one user strictly reduces the loss on another user, with influence decaying as the latent Wasserstein distance increases. We further study the induced geometry: with clean inputs, VAE-based CF primarily exploits emph{local} collaboration between input-similar users and under-utilizes global collaboration between far-but-related users. We compare two mechanisms that encourage emph{global} mixing and characterize their trade-offs: (1) $eta$-KL regularization directly tightens the information bottleneck, promoting posterior overlap but risking representational collapse if too large; (2) input masking induces stochastic geometric contractions and expansions, which can bring distant users onto the same latent neighborhood but also introduce neighborhood drift. To preserve user identity while enabling global consistency, we propose an anchor regularizer that aligns user posteriors with item embeddings, stabilizing users under masking and facilitating signal sharing across related items. Our analyses are validated on the Netflix, MovieLens-20M, and Million Song datasets. We also successfully deployed our proposed algorithm on an Amazon streaming platform following a successful online experiment.

Problem

• Analyzing how collaboration arises in VAE-based collaborative filtering systems
• Studying mechanisms to encourage global mixing between distant but related users
• Proposing anchor regularization to preserve user identity while enabling global consistency

Innovation

• Input masking induces stochastic geometric transformations
• Anchor regularizer aligns user posteriors with items
• Beta-KL regularization tightens the information bottleneck