Research On Recommendation Algorithm Based On Social Network

With the development of the web2.0technology, internet is playing a more and more important role in our daily life. While e-commerce is widely spreading among people, the issue as to promote users with useful information from the expending information has become a big concern. And thus Personalized service is more and more popular which provides the recommendation ability according to the habit and preference of the users.The content-based filtering and the collaborative filtering are the most successful personalized recommendation technologies which have been used in recommendation system. The content-based filtering is focus on the content of books interested by the users. The book profile and the user profile are represented in the same feature space. However, it has the semantic gap of modeling the book profile using the term frequency which does not reflect the content of book effectively. Collaborative filtering method predicts the rating of items for a particular user based on the ratings from other users with similar preference. Nevertheless, collaborative filtering has two serious limitations. One is the sparsity problem; the other is the cold start problem. The collaborative filtering commendation system is unable to make high quality recommendation to a new user with few activities and comments.The social net works such as Twitter, Douban, Facebook are becoming more and more popular nowadays, they formed rich relationships among users. And people in the social net can tag the item which they like. The relationships formed by interaction among users reflect the interest of user on one hand and also they represent the relationship among users. And thus it's a new approach to promote the recommendation result by getting useful information from these relationships and collaborative tagging.The main work of this paper is as follows:1. This paper proposes a new collaborative filtering algorithm based on collaborative tagging and community detection in the social network of the digital library. Firstly, the communities of the social network are detected to find the readers who have the similar preference. Then the user-book-tag triple model is built to calculate the best candidate tags. Finally, we use the naive Bayes classifier to recommend books. The proposed algorithm derived from collaborative tagging, which can solve the problem of the semantic gap. It is also able to settle the cold start problem attribute to using the community detection.2. This paper proposes a dynamic collaborative filtering algorithm based on collaborative tagging and dynamic interest transferring in the social network of the digital library. Firstly, the communities are partitioned in the social network of the digital library, each of which follows the assumption that similar tags are existed in the same social community. Then, the candidate tags are derived from the user-book-tag correlation model based on the dynamic user-based collaborative filtering algorithm constrained by the H.Ebbinghaus Forgetting Curve. Finally, the books with highest posterior of the tags are recommended by the naive Bayes classifier. The proposed algorithm derived from collaborative tagging and H.Ebbinghaus Forgetting Curve, which can solve the problem of the semantic gap. It is also able to settle the cold start problem attribute to using the community detection and the drifting of the users'interest.3. Based on the study of algorithms above, we implement the digital library platform based on community detection, collaborative tagging and dynamic interest transferring. We use Python and Flex to build recommendation system based on the three-layer model, so the users can enjoy reading service. Within three months, the recommendation system is the third one in the reading application of renren, which has two million users.