Research On Link Prediction Method In Dynamic Networks

Link prediction is a core issue in the research area of social network analysis, and predict the unknown part of network or the future network,by using the information obtained from the known part of network.Link prediction can be used to mine the missing information in the network,also can be used to predict what will happen in the future.It is widely used in the friend-recommendation system,co-author relation network and many orther fields.Real network has three characteristics:large-scale,sparsity and dynamic change.Obtaining useful information from large scale data is agreat challenge for link prediction;The real network has lots of nodes and few links,then how to use the information of pair of nodes not linked also become a problem;finally,the network is dynamic,in which nodes and edges are constantly updated,designing a dynamic link prediction method with the time characteristics in the network is a worth studying topic.The research content of this paper mainly includes the following three aspects:1 、This paper summarizes and analyzes the research status of link prediction.We make summarize the research in recent year releated with link prediction,and classify the research,then put out the main problem and challenge existing in the current work.This make a clear direction for further study of complex dynamic networks.2、 This paper proposes a semi-supervised link prediction method with the use of temporal metrics. Considering the sparsity problem in networks, a semi-supervised learning technique is used to exploit a large number of unlinked node pairs assisting linked node pairs in training process; Considering the network dynamics problem, several temporal metrics are used to describe node pairs. The experiments in two real datasets DBLP and Enron showed that the proposed method performed a higher prediction accuracy.3、Based on ensemble learning,we propose EnDLiP method. For the dynamics in the network, first,we select a number of structural features of the network.And then,we recorde change of every structural features in the evolution of network. A classifier is trained for each structural feature and a final ensemble result is obtained by weighting all the classifiers.