Research On Personalized Recommendation Algorithms Based On Network

With the rapid development of Internet technology, Internet products have obtained great popularity throughout the world. How to effectively utilize the user’s browse history for mining user preferences has become a big issue. Recommender systems aim at providing people things they interest in intelligently, and after decades’ development in this area, recommender technology had been successfully implemented in many disciplines.Traditional recommender systems manly focus on user object bipartite graph mining. Recently with the advent of social network service and social media, the relationships among online users require mining as well. In this paper, we propose a random walk based recommendation method in trust network. Meanwhile, we propose a diffusion-based nearest neighbor algorithm. The main work is list as below:1. Designed a trust network recommendation algorithm based on random walk. The algorithm can effectively both take advantage of the user object relationship and trust relationship between users. Experimental results show that the algorithm’s precision and recall rate are enhanced compared to the benchmark algorithm.2. Proposed a nearest neighbor recommendation algorithm based on random walk. We combined the traditional collaborative filtering nearest neighbor algorithm and mass diffusion method, and propose a random walk based nearest neighbor algorithm. Through the simulation experiments in the benchmark data sets, the algorithm can effectively improve the accuracy and diversity.3. This paper proposes a user interest modeling method based on text semantic analysis, and based on this method a collaborative filtering algorithm was proposed based on the user interest model. Meanwhile, as to mining unpopular things, we design a method to fuse heat spreading and tag based collaborative filtering. Finally, This paper proposes a logistic regression based fusion framework, which trains the parameters automatically. Through the simulations, the method proposed upon all enhanced the accuracy and novelty measures in recommender system.