How Do Recommendation Models Amplify Popularity Bias? An Analysis from the Spectral Perspective

Abstract

Recommendation Systems (RS) are often plagued by popularity bias. When training a recommendation model on a typically long-tailed dataset, the model tends to not only inherit this bias but often exacerbate it, resulting in over-representation of popular items in the recommendation lists. This study conducts comprehensive empirical and theoretical analyses to expose the root causes of this phenomenon, yielding two core insights: 1) Item popularity is memorized in the principal spectrum of the score matrix predicted by the recommendation model; 2) The dimension collapse phenomenon amplifies the relative prominence of the principal spectrum, thereby intensifying the popularity bias. Building on these insights, we propose a novel debiasing strategy that leverages a spectral norm regularizer to penalize the magnitude of the principal singular value. We have developed an efficient algorithm to expedite the calculation of the spectral norm by exploiting the spectral property of the score matrix. Extensive experiments across seven real-world datasets and three testing paradigms have been conducted to validate the superiority of the proposed method.

14 pages, 7 figures

References in corpus (19)

• DeepInf: Social Influence Prediction with Deep Learning
• Improving Graph Collaborative Filtering with Neighborhood-enriched Contrastive Learning
• Session-based Social Recommendation via Dynamic Graph Attention Networks
• Causal Intervention for Leveraging Popularity Bias in Recommendation
• A Survey on Accuracy-oriented Neural Recommendation: From Collaborative Filtering to Information-rich Recommendation
• Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System
• A mathematical theory of semantic development in deep neural networks
• Causal Embeddings for Recommendation
• Towards Representation Alignment and Uniformity in Collaborative Filtering
• AutoDebias: Learning to Debias for Recommendation
• Deconfounded Recommendation for Alleviating Bias Amplification
• News Session-Based Recommendations using Deep Neural Networks
• On the Effectiveness of Sampled Softmax Loss for Item Recommendation
• Alleviating Matthew Effect of Offline Reinforcement Learning in Interactive Recommendation
• Invariant Collaborative Filtering to Popularity Distribution Shift
• Balancing Unobserved Confounding with a Few Unbiased Ratings in Debiased Recommendations
• Debiasing Recommendation with Personal Popularity
• Curse of "Low" Dimensionality in Recommender Systems
• Popularity Debiasing from Exposure to Interaction in Collaborative Filtering

Cited by in corpus (1)

• Breaking the Top-$K$ Barrier: Advancing Top-$K$ Ranking Metrics Optimization in Recommender Systems