Potential Social Network Friends Recommend Algorithm

In recent years, social network(Social Network Service,SNS)is a novel, practical and convenient way to make friends, it depends on the character of authenticity and stability, obtains users’ favorite gradually.However, the development of the social networks faced with a phenomenon similar to the "Matthew Effect", the user scale of the more successful social networking sites are constantly expanding, and one of the most important purpose users using social networking sites is to get to know friends, maintain the relationship with friends, in order to expand their social circle, so when the social networking site having a huge user base, how to recommend friends to users is a problem of having to solve.Although scholars compare in-depth research on the friends recommend of social network, but the results are not perfect. Through observation and analysis in real social networking sites, we found that some users will gradually form a small group structure that is a community of a social network or you can understand it as the users’real social circle. Generally speaking, one user can only make friends with a few users in the social network, and the most likely users with that user to become friends are generally from the user’s social network, the probability of becoming friends with users from most of the other network is relatively low. In addition, on the realization of recommendation algorithm, most friend recommendation algorithm requires certain methods such as user link information subject and user preferences to construct the recommended model, and then produces recommendations by calculating the similarity, calculate the similarity in the social network in the hundreds of millions or even tens of millions of levels of large data sets, either in time or space overhead is huge and should not be ignored. In view of this, the algorithm of this paper adopted a kind of community identification algorithm for division of friends community, both narrowing the recommended range also reducing the size of the data for user similarity calculation. Just because at present most of the friends recommendation algorithm based solely on the link information between users to calculate the similarity of users, paid less attention to the relationship between the user and the strength of the relationship, this when making friends recommended, lost a lot of important information. In view of this, according to the division of community friends network diagram is established, and adding contexts of the real relationships to the edges in the graph, then a quantitative calculation of the strength of the relationship between users is conducted.Putting it and the similarity of users fusion, we put forward the concept of comprehensive user similarity used as the basis of a friend recommended.Finally, this paper uses the Java language to implement the algorithm and applies it to the friends data sets from the website of Renren we extracte for testing the validity and accuracy.