Research Of Personalized Recommendation Technology Based On Binary Trust Relationships

E-commerce recommendation system provides users with personalized recommendations timely and accurate services. So it can preferably serve customers and raise their loyalty and satisfaction to web site or enterprise, and to achieve the enterprise (site) and customer win-win good results.Trust has been introduced in e-commerce recommendation systems, and according to its social nature, trust between customers can be delivered, also the data sparsity and cold start etc. problems can be effectively alleviated. Systems match more neighbors to users, and get better recommended services for users. However, the existing expression for trust relationship between users is single and simple, it can not commendably simulate users’trust relationship in practice, and the available trust transfer mechanism should not show the actual transfer process well and truly. All those seriously affect the role of trust on e-commerce systems.This paper introduces the concept of trust support, and proposes a bi-trust relation model combining both trust and trust support. According to the social nature of trust, we propose the rule based on the principle that direct trust has precedence over the indirect trust, which has simulated the actual trust transfer process more preferably. We get the indirect trust between users by calculating the transitive trust,and finally get a binary customer trust network with the combination of indirect trust and direct trust.For the target this paper expects to achieve,and according to the established customer trust network, we design eight computing methods of the final weights and the corresponding recommendation algorithms. We also design three different experimental program based on the selected data set. Experiments show that, personalized recommendation method in this paper can improve the accuracy and coverage significantly. Experiments also prove that trust support extracted in this paper can also back up the recommendation method based on similarity.