Popularity bias reduction in multi-criteria recommender systems: the RelPref and PriRelPref approaches

Popularity bias represents a critical challenge in multi-criteria recommender systems, as collective trends may dominate individual preferences during personalization, thereby complicating the accurate modeling of users’ multi-dimensional preference structures. To address this issue, existing approaches for popularity bias mitigation are reviewed, and two novel methods, Relative Preference Model (RelPref) and Criterion-Aware Relative Preference Model (PriRelPref), are proposed. These methods model multi-criteria user--item interactions in a more precise and balanced manner by explicitly incorporating users’ criterion priorities. RelPref dynamically evaluates user ratings across multiple criteria, whereas PriRelPref further integrates user-specific criterion prioritization to more effectively control the influence of popularity. By aligning users’ most important criteria with the criteria in which items perform strongly, the proposed approaches generate recommendation lists that better reflect individual preferences while maintaining balance across criteria. Experimental results indicate that the proposed methods improve the overall quality of recommendations by enhancing personalization and supporting more balanced item exposure.