A Study On Collaborative Filtering Recommender Model Based On Trust

The rapid development of e-business brings people into the sea of commodity information. Customers can buy any kinds of goods around the world at anytime, anywhere when they want. Shopping becomes simple and easy, but with constantly development of network technology, the number of goods on the e-commerce increases exponentially. Eventually people have to face such huge commodity information, and they are troubled by"information overload"problems, though e-commerce makes their lives convenience.Therefore, how to help customers finding merchandise from such huge commodity information becomes the most important issue that the e-commerce site has to solve. As a way of personalized service, recommender system can effectively get user's interest, and recommends appropriated goods to them in order to meet their interests and hobbies, so it can help customer make decisions. Collaborative filtering is one kind of personalized recommendation technology, which is been used widely, but in this application process, there are several problems such as data sparsity, user trust and other key issues. However, because of the rapid development of web social network, trust-based personalized recommendation becomes one of the most important issues, because it brings trust mechanisms into traditional collaborative filtering recommender system, and can overcome shortcomings effectively as mentioned above.This paper focuses on how to enhance customer satisfaction with the recommendation provided by recommender system. In order to improve the traditional collaborative filtering technology, the management definition of trust is introduced into this collaborative filtering recommender system, which means that"trust is the expectation of technically competent role performance". More specifically, we interpret trust as one's expectation of another peer's competence in providing recommendations to reduce its uncertainty in predicting new item's ratings. So the ratings given by users on certain items can be fully used, for user matrix between two users is calculated based on this rating information. Besides, according to the conditional probability, a new trust calculation method is proposed. Based on this trust value, the active user's neighbor is formed. In the final stage, ratings provided by neighbors will be adjusted according to his trust value, and then this new rating value will be used in predicting items the active user probably would give. Combine the proposed trust calculation method and the characteristic of existing collaborative filtering recommender system, a framework of trust-based collaborative filtering recommender system is proposed, which can be used in the further development of this kind system. In addition, data from Movielens website is used to compare the traditional algorithm's and the proposed algorithm's performance in accuracy and robustness. The experiments'result shows that the proposed algorithm played better in this two metrics. Finally, the limitations and shortcomings of the proposed model and the future study are described in detail.