Study Of Link Prediction Based On The Features Of Nodes In Social Networks

With the development of computer technology,the network technology is changing our life.Many kinds of popularity social networks and convenience intelligent terminals make people leave a mass of real date in the virtual network.It could provide a data basis for network analysis researches.Social network is becoming greater over time.When we are enjoying our colorful life,we also fell many annoyances of information expansion.It's difficult for us to find some relevant information quickly and efficiently in huge amounts of data.Predicting the possible of new links and unfounded links,will advantage to find some lacking data and analysis the problem of the complex network evolution mechanism.It's significant to the researches of the social networks.How to predict new links effectively base on the existing information is the fundamental problem of network analysis,and in popular commercial social networking sites field,it was also demanded widely now.At present,there are three kinds of link prediction algorithms: local similarity-based algorithm,path-based algorithm and random walk-based algorithm.Similarity-based algorithms are simpler and low time consuming was widely used.The other two algorithms are complex and time consuming,which are usually infeasible for large-scale networks.The existing similarity-based prediction algorithms were not making full use of the information of nodes,results in decreasing of prediction accuracy.Improving the accuracy of link prediction is one of the basic problems for the complex network researches.In this dissertation,we study the problem of link prediction in social network,and the main work and contributions are as follow:1.Some important nodes may have a greater influence or stronger ability in information dissemination.The existing similarity-based prediction indexes were not making full use of the influences of nodes.In order to solve the above problem,a link prediction algorithm which basing the importance of nodes was proposed.In this method,it used the network degree centrality,closeness centrality and betweeness centrality on the basis of common neighbor and Adamic-Adar,resource allocation indexes of similarity-based prediction method.The prediction indexes of CN?AA and RA with considering the importance of nodes were proposed to calculate the node similarity.Experiment was on four real-world networks and AUC standard metrics was adopted in this dissertation,and the experimental results show that the accuracy of link prediction on four data set is higher than comparison algorithms,which means the proposed approach outperforms the traditional link prediction algorithm.2.By mining the deep interaction between neighbor nodes,we proposed the algorithm of node clustering ability.This algorithm could filter some topology information of network carried by the common neighbors.3.The existing similarity-based prediction indexes were not making full use of the clustering ability of nodes.In order to solve the above problem,a link prediction algorithm basing on the clustering ability of nodes was proposed.The clustering information of common neighbors was fully used by the new algorithm,and the clustering ability of nodes plays a more important role in the process of link prediction.Experiment was on four types of real-world date sets,based on the MATLAB simulation tools,and the AUC and Precision indexes were adopted in this dissertation,and the improved algorithm can generate more accurate prediction results.