Incorporating Social Networks and User Opinions for Collaborative Recommendation: Local Trust Network based Method

The sparse nature of historical rating profile hinders reliable similarity metrics between users, leading to poor recommendation performance. The availability of user social networks and user opinions can be incorporated to improve prediction accuracy. One of the key points is how to make the multiple sources of information consistent for the purpose of recommendation. In this paper, we proposed Local Trust Network (LTN) based recommendation method in the setting of movie recommendation, that mines the social network and multiple sources of user opinions to generate a highly reliable trust user network, upon which a recommendation is made. With transductive reasoning, LTN interpret trust user as a collection of instances, so it is well suited for the sparse issue of social network information. Our experiments on CAMRa10 data set shows the proposed methods improve recommendation performance significantly.