Research On Social Network Link Prediction Model Based On LDA

With the fast development of Web2.0technology, the services based on the social community are more and more popular, for example, Wikipedia, Flickr and Facebook. In the community, people can find and upload their favorite photos, link to other users. Social Networks are the systems that relying on the friendships between people, they are a subset of interpersonal networks. The number of social network users has increased dramatically and it has brought great challenges to mining the relationship of the users, the information of the users in the network. One data mining problem of interest for social networks and the characteristics of the structure for social networks is the friendship link prediction problem.However,many social network link prediction algorithms focus on the topology structure similarity between nodes in the networks, for example,CN algorithm[1], A A algorithm[2],katz[3] algorithm and so on.But these algorithms don’t focus on the semantic relations between user’s interests.lt results that the accuracy of link prediction is not high.At the same time, the user interests in the social network are huge, if we analysis the semantic relationship of them directly, we will spend a lot of time.To solve the above problems,we introduce the topic model,the Latent Dirichlet model.First,we use LDA to model the interests between the user nodes and extract the topics of users’interests in the social network,analyze the semantic similarity between these topics. Compared with the direct analysis of interests for each user,it not only can capture the latent semantic relations,but also can reduce the computation time complexity effectively.Then for the first time,we use the Rescource Allocation algorithm in the complex network into the social network to grasp the structure feature accurately.At last,by constructing the classifiers,we use of supervised learning framework integrate semantic feature and the toplogical structural characteristics to predict the friendship links in the social network.We apply the Social Network Link Prediction Model based on LDA to a real social network called LiveJournal, compare with the other methods to verify its viability and effectiveness. Experimental results on the subset of LiveJournal show the usefulness of the LDA features and structure features for predicting friendships.