Research On Friend Recommendation Based On Community Detection

With the rapid development of information technology, the Social Network Service, one of the most popular tools for making friends, plays a more and more important role in human society. As a major platform for building friendship and sharing interests online, SNS benefits users a lot. Users can express their ideas every time and share their interesting things with their friends through using SNS tools. What's more, users can also make friends with people who have the same interesting. But there are more and more registered users, which can generate tremendous messages each day. This scale benefits the uses but it can also flood users with huge volumes of information and hence puts them at risk of information overload, so it is difficult for users to make friends who they want to make. In order to solve this problem, people make various ideas, including the idea of making models, discovery community. But there are many existing problems such as the method people they use now to discovery community is not perfect. People usually don't take the weight in graph into account when they discovery community which can draw the wrong conclusions. What's more, the scale of complex network become more and more tremendous large. So the most important task which needed to be competed now is how to use the big data to make services to people.In this paper, a community discovery algorithm is proposed which based on label propagation and used the weight in graph. First of all, a new concept called link strength is composed, which can be calculated through the interactions of people who both make friends with each other. Then the link strength is used to be the weight of the weighted network which can be got by the real society.Secondly, in order to improve the accuracy of community detection, the existing label propagation algorithm is improved. We take the weight into account when we detect community. In order to make more accuracy friends recommendation, we first find the community where the target user belongs to, then calculate the similarity between the target user and the people who are in the same community with the target user and they both not make friends with each other now, finally the Top-k users who are the most similar to the target user are recommended to the target user.Then, by utilizing the Map-Reduce framework on Hadoop, a distributed scheme of the method is proposed and implementation.In the finally, it is shown that our method performs better than the other methods of friend recommendation with higher accuracy and stability on both synthetic and real networks which is proved by experiments. It is also proved the distributed scheme works well on large weighted complex networks.