Research On Link Prediction Method In Dynamic Networks

With the popularity of the Internet,the network data has an extraordinary growth.How to find useful information from the researchers become increasingly important.As an important branch of link mining,link prediction can be used to provide valuable information for the people by modeling and analyzing the known network.The real network has two characteristics.On the one hand,the network is dynamic.With the passage of time,the number of nodes and links in the network are constantly updated.The time information of nodes forming links can usually reflect the potential relationship between nodes.On the other hand,although the number of nodes in large scale networks is large,the number of links between nodes is relatively small.If we can choose some representative samples,we can reduce the training pressure and maintain a better prediction effect.This topic mainly through the following three aspects of research:1.T This paper summarizes some famous link prediction methods in recent years and puts forward the bottlenecks and challenges of the current link prediction tasks.2.According to the characteristics of network dynamics,this paper proposes a dynamic link prediction method based on ensemble learning,which is called Dynamic.Traditional link prediction methods are based on static network structure prediction of hidden links,while ignoring the potential of information network in the dynamic evolution process.This paper puts forward the method using machine learning technology to change the network structure for training.3.Aiming at the characteristics of network sparsity,this paper introduces an active learning paradigm in the process of network evolution and link prediction,which is called DynActive.A classifier generation method for each sequence of the changes in the characteristics of the network structure,and then use the classifier to each node connected to the node score prediction results of different samples submitted by the users,once get real mark system is used to update the training set to train the classifier and integrated into the final model.The experimental results of three co-authors network data sets show that the AUC index has been significantly improved by introducing ensemble learning and active learning in the dynamic network link prediction method.