Link Prediction By Neighbor Communities And Node Importance

With the development of internet,complex networks have been drawn more and more attention.As one of the most important topics of complex networks,link prediction is of both theoretical values and practical applications in several fields.Four link prediction algorithms are proposed,which are based on the information of neighbour communities and node importance.The main contributions are listed as follows:(1)based on the algorithms of hard community detection,we proposed an algorithm of soft community detection,and then gave a link prediction method by statistical inference;(2)based on the technique of maximum likelihood estimate,a bi-scale method of link prediction is proposed to define the probability of a link between two nodes,which fuses a microscale information(neighbor)and a mesoscale information(community);(3)based on node global importance(node betweenness),a link prediction method is proposed for relative small networks;(4)a link prediction method is proposed to fuse the network topological information(node degree)and geometrical information(distances between nodes,which is obtained by force-directed placement).Experimental results indicate that: the first method performs better than those based on hard communities,which indicates at certain levels that the soft communities are more applicable to real networks;the second method performs better than those based on single scale information,which potentially indicates that the multiscale characteristic of the connection mechanism of real networks;the last two methods(which are based on global importances of nodes)also perform better than some typical methods based on local information(neighbor)which also indicates that the real shortest paths or potential distances(obtained by global optimization)between nodes are also responsible the evolution of networks at certain degrees.