Clustering based transfer learning in cross domain recommender system

Abstract

The technique of collaborative filtering in recommender system suffers from data sparsity and cold start. In this paper, a cluster based approach is proposed for alleviating the problem of sparsity by transferring the knowledge from a more densely rated concomitant domain. The paper focuses on providing recommendation in a sparsely rated domain by transferring the knowledge from the highly rated domain with the same users rating the items in both the domains. This cross domain system transfers the affinity among users from the highly rated domain to the sparsely rated domain to make more accurate recommendation at the target domain. An algorithm is proposed to link the users' affinity between the domains. The proposed cross domain algorithm is tested with various clustering methods. The experiments are performed considering restaurant — tourist attraction as cross domains and results show that hierarchical agglomerative clustering performs better when transferring user affinity between associated domains.