Novel approaches to measuring the popularity inclination of users for the popularity bias problem

An efficient approach to handling the well-known popularity bias in recommendations is treating this issue by considering users' actual propensities on item popularity. This way, the recommender systems can also provide more calibrated individual recommendations. However, when estimating users' tendency on popularity, the existing methods consider all rated items in their profiles, including even negatively rated ones. In this study, we propose two novel approaches, namely Better-Than-Average (BTA) and Positively-Rated (PR), to measure the popularity inclination of individuals more correctly. Both methods aim to consider only items favored by users rather than all rated ones but differ in how they select the set of liked items. More specifically, the BTA filters out items with higher ratings than the average rating value of the user profile, while the PR considers only positively-rated ones based on the rating scale. The experiments conducted on three benchmark datasets conclude that our approaches measure the level of users' popularity inclination highly different when compared to existing methods, and such differences reach significant degrees in some cases. We also observe that the proposed BTA is more sensitive than the PR in measuring popularity inclination due to its followed mechanism for forming the set of liked items. We also conclude that the impact of our BTA and PR approaches on popularity inclination is more distinct for highly-engaged users, who are the most critical stakeholders of a recommender system.