Studies On Collaborative Filtering Recommendation Based On Trust Network And Random Walk

With the rapid development of the Internet and information technology in recent years, more and more information have been provided to user by network. When the Internet brought the convenient to us, the problem of information overload was becoming progressively problematic. Now, finding a way to get useful information from massive data is necessary to people. Many web applications, such as search engine(Google, Baidu), portal sites and data indexing system, are a kind of tools which help people to screen and acquire information, but they cannot make people satisfied.As a kind of potent information filtering technology, personalized recommendation system is an important method to solve the phenomenon of information overload. Collaborative filtering, which has been developed greatly in theory research and practical application, is one of the most widely used techniques for recommendation system which has been successfully applied in many applications. However, it suffers from serious cold start, sparse data and fake info attack because of the increasingly number of users and dataIn order to fix the above problems and improve the precision of recommend system, the main work what this article has done as follows:(1) Firstly, this article has reviewed and analyzed the traditional collaborative filtering recommendation algorithm. According to the features of trust relationship in web, this article has constructed a trust network.(2) Secondly, in this paper, we improved the random walk model combining trust-based and item-based collaborative filtering method for recommendation. The trust factor is considered as an important measure of guiding recommendations. The random walk model considers not only ratings of the target item, but also those of similar items. The probability of using the rating of a similar item instead of a rating for the target item increases with increasing length of the walk. At last, we will calculate the consequences of each round weightily to get the final result.(3) Thirdly, this article combines the traditional Pearson similarity method and Sigmoid function. The improved method has raised the influence of common rated users.(4) The empirical analysis on the Epinions dataset demonstrates that our method can further improve in terms of MAE relative to other traditional collaborative filtering algorithms.