Research On Link Prediction Of Social Networks

The data on Internet show the explosion of growth in recent ten years.How to dig outthe useful information and choose the accurate information the user need has been a hotresearch direction in data mining. Sometimes we need to make a prediction of futureinformation and its development trend on the basis of the existing information. Linkpredicton arises at the historic moment. The link prediction problem in the network isusing the known network structure and related information, such as people’s connection toeach other of sina microblog, to predict the possibility of connection between two nodesand to predict the probability of two nodes’ association in the future. Predicting thoseexisted but not known links is actually a data mining process and predicting those linkswhich is likely to appear is estimating the evolution process of a network.There have been mainly three traditional link prediction methods. The first one kindis based on the markov chain, supporting vector machine and machine learning. Thesemethods just consider the node attributes feature. Although they can get a high predictionprecision but its application scope is very narrow, which is only applicable to networkswhose node attribute are real and clear. The second one kind is based on the maximumlikelihood estimation of network structure, which mainly take the topology of networkinto consideration and ignore the attributes of network nodes. But it also have the highcalculation complexity problem. The third kind is based on the similarity of nodes in thenetwork and it has a satisfied prediction accuracy, but its performance can be better.This paper proposes and improved node similarity judging algorithm, which fullyconsiderd the network topology, introduced more node similarity indexes and joined theimportant index of node attributes. Considering the special sina micro blog service, thispaper put forward a novel kind of hybrid time series link prediction algorithm to improvethe node similarity algorithm. This algorithm considers network structure of many timepoints and forecast the future links by modeling the network. This method can not onlyenhance thd prediction accuracy, especially for social network, but also has a lowcomputational complexity.