Researches On Recommendation System Based On Trust Diffusion Mechanism

In recent years, with the comprehensive popularization of the Internet and the rapid development of electronic commerce, recommendation systems has become a hot research topic. Recommenders intelligently recommend resources users are interested in to users by studying the users’interests and characteristics of information, which can help users save a plentiful of time, and help businesses create a lot of value simultaneously. Recommendation systems have been widely noted for their many advantages, and get great development in terms of theory and practice. However, because the rapid development of network causes a surge of online information, the traditional recommendation systems cannot satisfy the needs of users. They exist many shortcomings, including the inaccuracy, data sparsity and lack of trust in the process of recommendation.Under the situation of rapid development Internet, the personalized recommendation systems based on trust introduce the trust mechanism in traditional collaborative filtering recommendation, which can effectively overcome the above defects and improve the performance of the recommendation system, thus become one of the current important research problems. However, the recommendations based on the trust are still a few questions:firstly, the sparse of trust network. In recommender systems, the number of users is often very large, but in the process of practical recommendations, the number of direct transactions between the users is small. So the number of direct trust relationships established by the limited number of transactions is also very little, and the number of the users directly associated with the target users is also relatively little, which leads to the direct trust relationships can only play a small role in the process of the recommendation. Many recommendation algorithms based on trust only use the direct trust relationships of trust network, do not consider indirect trust relationships. Secondly, the dynamics of the trust network. In the trust network, the trust relationships between users are not static or invariable, but dynamically change over time, or other events, the change of the trust relationships may result in changes of recommendation results. Therefore, considering the dynamic trust network has a great influence on the recommendations is also an important research direction. Thirdly, how to integrate the trust model with the recommendation system. Most recommendation algorithms based on the traditional probabilistic matrix factorization model, fusion between user-item matrix and social relationships by sharing a potential low dimensional user characteristic matrix. This method can only learn few effective characteristics, which can not reflect the recommendation process in real life.This study aims at handling the above problems in process of recommendation. On the basis of existing research, this paper presents a new recommendation method based trust diffusion mechanism (DiffTrust+RSTE).(1) According to the diffusion theory in economics, we think the trust as a dynamic process, namely the trust may change as time and interactive situation changes. We also consider the factors such as the heterogeneous characteristics of both spatial and temporal information, and time decay factor. On the basis of these, we design a new trust diffusion model (DiffTrust) suitable for collaborative filtering recommendation systems. DiffTrust makes full use of the direct trust relationship between users, to derive the indirect trust relationship between users through a certain trust propagation rules.(2) We employ the social trust ensemble (RSTE) to put DiffTrust into the process of the recommendation. DiffTrust+RSTE has the following characteristics:On one hand, this method matches more trust users for the current user through DiffTrust model, fully digs up more new trust relationships between users and used for recommendation service, which perfectly solves the data sparse of the trust network. On the other hand, DiffTrust+RSTE combines the users’tastes with their trusted friends’interests through a set of parameters. This consideration can truly reflect the recommendation process in real life.(3) We evaluate the DiffTrust+RSTE algorithm on Flixster, Moviedata and Epinions datasets, respectively. The experimental results show that the recommendation based on our proposed DiffTrust+RSTE achieves high performance in each evaluation metric (RMSE, Precision, Recall and FMeasure). Finally, the limitations and shortcomings of the proposed model and the future study are described in detail.