A new similarity measure to increase coverage of rating predictions for collaborative filtering

Abstract: User-based collaborative filtering (UBCF) is a popular technique in recommendation systems, where rating estimations for unrated items are based on the rating information of each neighboring user. However, conventional similarity measures for identifying the neighboring users often overlook the user’s contribution in rating predictions for unrated items. This leads to limited coverage in rating predictions. This study addresses this limitation by introducing a novel similarity metric that effectively incorporates the degree of user contribution to unrated item predictions, thereby enhancing coverage. Additionally, the proposed approach assigns varying weights to items based on their rating frequency, further improving coverage. Extensive experiments were conducted on six benchmark datasets to validate the proposed approach. The results showed that the proposed similarity measure improved the recommendations and significantly expanded the coverage of predictable items. Furthermore, the proposed similarity enhances diversity in recommendations, contributing to a more robust and effective recommendation system. Graphical abstract: [Figure not available: see fulltext.]