The Optimization Study Of Matrix Factorization Based The Transitivity Of Social Trust

With the WEB technology providing new energy to development of Internet, it leads to huge evolution in social network, encyclopedia system and e-ecommerce system. The main feature of these applications is focusing on user and data-driven. The users play an important role as data consumer. Meanwhile, they produce data. Huge information, which uses to meet all demands of different kinds of information, brings unprecedented opportunity since the information on the Internet is increasing dramatically. However, some rigorous challenges of information processing technology are produced, which mainly refers to'information overload'. However, the information filtering technology represented by Recommended system can provide better method to resolve this problem. The principle of filtering technology is excavating a large number of users behaviour data, analysing users requirements and then pushing more interested information to users. Collaborative filtering is one of the most extensive and successful algorithm. It can be widely used for different kinds of recommended scenes and it only depends on the users historical behaviour data. The neighbourhood approach and Latent Factor models are the most two successful algorithms in collaborative filtering. The most successful Latent Factor model is Matrix Factorization. The core concept of collaborative filtering is using the historical behavior data of users, finding specifying similar members of users from user groups, synthesizing the information evaluation from these similar users and then forming the degree of preference. However,the collaborative filtering faces two serious problems: data sparsity and cold start.With the development of online social network, it plays more significant role in collaborative filtering. During recent years, it is widely studied and used in recommended system, which has made obvious progress. In this article, it is focusing on the transmissibility of social trust network creates a new social trust matrix and proposing the matrix factorization model of social trust transmissibility. Furthermore, the contribution of this article is raising a new trust prediction model in order to obtain a new trust matrix, which contains the direct and indirect trust information among users. The new matrix replaces the original one to integrate into matrix factorization,which results to create a new matrix factorization based on social trust. The article adopts a real dataset to verify algorithm. The results indicate it is effective to relieve data sparsity and cold start. Moreover, it decreases recommend error and increases the prediction accuracy and recommend quality.