Leveraging Transitive Trust Relations to Improve Cross-Domain Recommendation

Abstract

Cross-domain recommendation has become increasingly popular because it can determine dependencies and correlations among different domains. However, this type of recommendation still suffers from data sparsity limitations. Social trust relationship helps alleviate this problem. Existing cross-domain recommendation algorithms focus on modeling user behavior in different domains but disregard the social trust relationship among users. In this paper, we propose a transitive trust-aware cross-domain recommendation model that incorporates the context dependence and transitivity of social trust relations. First, we construct a different transitive trust network for each single domain. Then, we develop a novel probabilistic matrix factorization model for each domain that utilizes the transitive trust-aware model to mine social trust. Finally, we present a nonlinear user-vector mapping algorithm to bridge the feedback of different domains. Experimental results indicate that our method significantly outperforms several state-of-the-art methods, produces higher rating prediction accuracy, and exhibits better recommendation performance on several real-world recommendation tasks.