Abstract: Personalized recommender systems often focus excessively on popular items when generating recommendation lists, ignoring the user's need for relevance and diversity in the recommendation set. Therefore, eliminating the popularity bias in recommender systems has become a widely concerned problem. Although state-of-the-art causal inference methods have revealed the complex connection between user-item interactions and popularity and achieved some success in mitigating popularity bias, existing studies often fail to model the temporal dynamics of popularity distributions, which leads to incomplete debiasing and, consequently, affects the diversity of recommendation results. Therefore, an innovative debiasing mechanism, modeling popularity and quality preference of users (MPQP) is proposed, aiming to address the impact of popularity bias on recommendation models. Specifically, MPQP introduces time-separated item popularity distributions in the causal inference process to capture users' preference for popularity. In addition, MPQP incorporates item quality distributions to model user preferences for quality. Experimental results show that MPQP significantly outperforms current state-of-the-art benchmark methods in terms of performance.