Research On Personalized Recommendation Service Based On Social Network

With the rapid development of information technology and social network, to satisfy the growing individual demands of customers on social network is becoming an urgent problem to settle down. Therefore, the study on the personalized recommendation in social network is of great theoretical as well as practical significances. This thesis is written based on the cooperation project of a software company in Xi’an and a U. S. Internet company, studies classic community detecting algorithms, personalized recommendation technology and trust evaluation method based on social network, and proposes an improved personalized recommendation algorithm based on social network. The effectiveness of this algorithm is finally demonstrated by experiment. The main job of this thesis is as following:1. Analyze and summarize the characteristics and theoretical foundation of social network. Introduce some classic algorithms of community detecting in social network research and analyze the advantages and disadvantages of the community detecting methods. Propose an improved community detecting algorithm based on the phenomenon of "stars" and "fans" in real life.2. Some commonly used personalized recommendation algorithms are introduced, including information retrieval technology, association rules, content-based recommendation technology and collaborative filtering technology. User-based collaborative filtering recommendation and item-based collaborative filtering recommendation are emphatically researched.3. Based on the characteristics of social network and shortages of the traditional collaborative filtering algorithm, this thesis puts forward an algorithm based on social network. Firstly, the improved community detecting algorithm is used to cluster users based on interests in this algorithm, and select candidate neighbors set fast, reducing the time of recommender system. Secondly, simulate the trust evaluation method among persons in the social network to calculate the trustworthiness between users, which is used to have a father screening for the nearest neighbors set. Finally, predict the score of a user on the items never scored based on the trustworthiness to recommend, making up for the deficiency of the traditional collaborative filtering, which is magnifying the similarity in the process of recommendation.4. Finally, this thesis fetches the experiment data set through the API on Flixster, which is a movie social network, to conduct the contrast experiments between the traditional collaborative filtering based on Pearson correlation coefficient and the improved algorithm. Results demonstrate that the recommendation accuracy and speed of the algorithm proposed in this thesis is superior to the collaborative filtering algorithm based on Pearson similarity.