This page contains supplementary information for the paper "Hybrid Recommendations by Content-Aligned Bayesian Personalized Ranking" by Ladislav Peska.
In many application domains of recommender systems, content-based information are available for users, objects or both. Such information can be processed during recommendation and significantly decrease the effect of the cold-start problem. However, content information may come from several, possibly external, sources, which may be incomplete, less reliable or less relevant for the purpose of the recommendation. Therefore, each content source possess a different level of informativeness, which should be taken into consideration during the process of recommendation.
In this paper, we propose Content-Aligned Bayesian Personalized Ranking Matrix Factorization (CABPR). CABPR aims to incorporate multiple sources of content information into the BPR. The working principle of the proposal is to create user-to-user or object-to-object similarity matrices based on the content information and impose similarity of latent factors of closely related users’ and objects’. CABPR also estimates relevance of similarity matrices during the training phase. Several variants of CABPR were evaluated on two datasets: MovieLens 1M dataset, extended by the content information from IMDB, DBTropes and ZIP code statistics and LOD-RecSys (books) dataset extended by the information available from DBPedia.
Experiments shown that proposed methods significantly improves over standard BPR as well as previously published BPR_MCA extension of BPR under several cold-start scenarios.