Research On Link Analysis And Prediction In The Social Network

With the development of Web2.0and hundreds of millions of participants, socialnetwork services based on community, such as Twitter and FaceBook have becomecommonplace. At the same time, The network link relations also gradually complicated, linkmining has become a research hotspot. Especially, as an important branch of link mining-theimportance of the link prediction has become increasingly prominent. A variety of activities inthe online community such as the user community development and looking for the topics ofinterest all need to be supported by link prediction technique. The existing social networklinks prediction method has a lot of problems, Such as node attribute information and networktopology information is difficult to comprehensive consideration, the accuracy of predictedresults also be enhanced urgently. In order to solve this problem, this paper puts forward twosocial network link prediction methods based on LDA model and PropFlow algorithm, andthrough the Supervised learning framework, form a joint prediction model, trying to achievethe goal of comprehensive consideration the multiple information of social network, and toimprove the accuracy of link prediction. Specific work of this paper mainly includes thefollowing aspects:First of all, this paper analyses the existing social network link prediction methods,summarizes the advantages and disadvantages of each method, then finding out thelimitations of existing link prediction model.Secondly, in response to the problem of existing algorithms lack the semantic relation ofuser attribute information mining, we proposed social network link prediction method basedon the LDA model, the method use LDA model to extract the topics between the socialnetwork users, then analyze the semantic similarity between these topics, and calculate thelink prediction accuracy.In addition, on the basis of these two social network link prediction methods, weproposed a joint prediction model. The model through constructing the classifiers, we use ofsupervised learning framework integrate semantic feature and the toplogical friendship linksin the social network.Finally, This paper designs and achieves the experiments, and verifies the feasibility andaccuracy of the above method. We adopted the NLPIR data sets which is provided by websearch mining and security lab. of Beijing Institute of Technology to experiment.we comparethe PropFlow algorithm, Parimi method with the method presented in this paper and analyze and discuss the experimental results. Experimental results show that the method of the papercan effectively improve the precision of link prediction.