Taming Recommendation Bias with Causal Intervention on Evolving Personal Popularity

2025-05-20

AI Summary

Popularity bias—where popular items are over-recommended—pervasively undermines recommendation accuracy and fairness. Existing debiasing methods predominantly employ static, global strategies, neglecting both individual heterogeneity in users’ popularity preferences and their temporal evolution. To address this, we propose “evolving personal popularity,” a novel metric that captures the time-varying nature of each user’s preference for popular items, and construct a causal graph incorporating this dynamic variable. We further design a deconfounded training framework that jointly models user–item co-evolution and performs causal intervention within the dynamic graph. Extensive experiments on multiple benchmark datasets demonstrate that our approach significantly mitigates popularity bias while improving both recommendation accuracy and group-level fairness. To the best of our knowledge, this is the first work to enable temporal causal modeling and intervention at the user level for popularity preferences.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Popularity bias occurs when popular items are recommended far more frequently than they should be, negatively impacting both user experience and recommendation accuracy. Existing debiasing methods mitigate popularity bias often uniformly across all users and only partially consider the time evolution of users or items. However, users have different levels of preference for item popularity, and this preference is evolving over time. To address these issues, we propose a novel method called CausalEPP (Causal Intervention on Evolving Personal Popularity) for taming recommendation bias, which accounts for the evolving personal popularity of users. Specifically, we first introduce a metric called {Evolving Personal Popularity} to quantify each user's preference for popular items. Then, we design a causal graph that integrates evolving personal popularity into the conformity effect, and apply deconfounded training to mitigate the popularity bias of the causal graph. During inference, we consider the evolution consistency between users and items to achieve a better recommendation. Empirical studies demonstrate that CausalEPP outperforms baseline methods in reducing popularity bias while improving recommendation accuracy.

Problem

• Mitigating popularity bias in recommendations by considering evolving user preferences
• Addressing time-varying personal popularity to improve recommendation accuracy
• Proposing causal intervention to reduce bias while maintaining recommendation performance

Innovation

• Introduces Evolving Personal Popularity metric
• Applies deconfounded training on causal graph
• Considers evolution consistency during inference