Research On Link Prediction Algorithm Based On Mutual Information And Node Centrality

Link prediction is one of import,ant research directions in complex networks,which has significant meanings in both theoretical research and practical applications.Most research works mainly focus on static networks,which neglect time evolution information of networks,and many link prediction algorithms ignore the differences of common neighbors,so there are certain deficiencies.This thesis mainly focuses on following aspects:1,how to distinguish different neighbors effectively in link prediction;2,how to perform link prediction in dynamical networks with.history information;3,how to consider the effects of node centrality in link prediction.The main works are as follows:1?Research in link prediction algorithm based on mutual information in static networks.In order to distinguish different neighbors,a modified mutual information(MMI)algorithm was proposed with common neighbors' degree based on traditional mutual information(MI)algorithm,MMI not only considers structural information about common neighbors,but also distinguishes different neighbors with degree information.The result shows that degree information of common neighbors plays an important role in link prediction in static networks.2\Application of modified mutual information algorithm in dynamical networks.In order to make full use of history information of networks,time series model was used to represent dynamical networks,and moving average modified mutual information(MA_MMI)algorithm was proposed by combining modified mutual information algorithm with moving average model.MA_MMI not only considers structural information and degree information,but also considers the effect of history information.The result shows that MA_MMI can achieve better than MMI in precision when the network is not sparse.3?Study the effects of node centrality in link prediction in dynamical networks.Considering the discriminations between nodes,the importances of nodes were identified first by node centrality methods,then combined with link prediction algorithms which already exist after normalizing.The results in four dynamical networks show that link prediction algorithms with node centrality can achieve better performance in dynamical networks.