Research on mitigating popularity bias in federal recommendation based on users' behavior

In recent years, with the enhancement of user privacy awareness and government regulations on privacy protection, the combination of personalized recommendation technology and privacy protection technology has become a trend. Although federated recommendation technology effectively addresses the issue of user privacy leakage, our research has found the phenomenon of popularity bias in federated recommendation, where popular products are prioritized for recommendation. This phenomenon results in unfair competition among products and affects e-commerce platform development. It is imperative to address the issue of popularity bias in recommendations. In this paper, we study the issue of popularity bias in recommendation systems under the federated learning framework. First, we quantitatively analyze the popularity bias in federated recommendation models, demonstrating the presence of strong popularity bias in their recommendation results. Secondly, drawing upon psychological theories and considering the impact of social groups and exposure effects on user behavior, we explore the behavioral influences contributing to popularity bias and design a debiasing method suitable for federated recommendations. Finally, we propose a strategy for mitigating popularity bias within the federated recommendation framework, which can simultaneously address both privacy protection and popularity bias, tackling these two significant issues. We validate the effectiveness of our debiasing method on two publicly available datasets, achieving data security while reducing popularity bias and improving recommendation accuracy.