Tags and Item Features as a Bridge for Cross-Domain Recommender Systems

Collaborative filleting is one of the widely implemented techniques in the area of recommender systems. But it suffers from data sparsity problem. To address that problem, cross-domain recommender systems (CDRSs) have been emerged to solve the data sparsity problem and improve the accuracy of prediction by transfer learning mechanism. To apply transfer learning mechanism, some common properties associated with users and/or items are needed between the domains. Several attempts have shown that recommendation quality of cross-domain recommender systems could be improved by transferring the user-generated tag information into the target domain. However, sometimes that information is not enough to accomplish recommendation task efficiently. To this end, item features can also be a valuable source of information for developing the correlation between domains and would be considered in generating effective recommendations in target domain. In this paper, we propose a model by utilizing item features and user-generated tags through matrix factorization in CDRSs framework. Firstly, we extract item features in terms of genres and user preferences in terms of user-generated tags. Thereafter, to establish the bridge for transferring knowledge, matrix factorization has been used. Finally, experimental results demonstrate that our proposed model outperforms the other single domain as well as cross domain approaches in CDRSs framework. (C) 2018 The Authors. Published by Elsevier B.V.