Friends Recommendation Based On Graph Ranking On Social Network Site

With the rapid development of web2.0 technology, the global Internet goes into the SNS era. A variety of SNS sites are emerging, such as Facebook, twitter, Myspace, Flickr and so on. In these sites, users can add other users as friends. For a user, how to find friends for him becomes a difficult problem. Friends recommendation system is the emergence of this problem on social networking sites.In this dissertation, we analyze friends recommendation mechanism, and propose two sets of issues:hot friends recommendation and personalized friends recommendation.In hot friends recommendation section, we propose the concept of user's behavior patterns in social networking sites. Firstly, we describe the user behavior patterns from three aspects:the user profile creation, content creation and relationship building. Secondly, we propose a model of social network siting surfering. And, we design an evaluation indicators to measure the user's hot degree. Hot users are recommended with high probability. Finally, we analyze the distribution of the users' information propagation contribution in the experiment. It was found that the distribution of hot degree is the power-law distribution. The results show that most users have relatively small hot degree, and only a few users have the relatively large hot degree.In personalized friends recommendation part, we use three algorithms for personalized friends recommendation problem. The three algorithms inlcude the naive algorithm, Personalized PageRank algorithm, and Multi-type Interrelated Objects Embedding(MIOE) algorithm. We anaysize the effect of friends recommendation of three algorithms in experiments. The experimental results show that MIOE algorithm is better than the first two methods. This is mainly because MIOE can use more user information.