| In recent years,the research on complex networks is developing rapidly.As one of the major research objects for the complex network,the link prediction takes the prediction of the possibility of making links among the nodes,which have not formed the link in the network,as a main research task.The research on link prediction is not only of profound theoretical significance,but also can be widely used in the fields of commodity recommendation,biological medication and convenient traveling,which effectively improves people’s living standard.Traditional prediction algorithms are mainly based on the number or the degree value of common neighbor nodes to predict the possibility of linking among nodes.It ignores the impact of nodes with important information on the similarity transmission,and pays little attention to the long paths.Therefore,this paper proposed new predicting indexes based on the importance of nodes and the heterogeneity of link paths.The main works of this paper are as follows:Firstly,the concept of “Node Importance” is proposed based on the degree value and centrality index of nodes.And adaptive parameter is set to adjust the proportion of each value in the node importance index according to the network differences.Then,based on the node importance index,an Optimize Clustering Coefficient(OCC)index,an Optimize Betweenness Centrality(OBC)index and an Optimize Closeness Centrality(OCL)index have been put forward.The experimental results show that the new indexes have better predictive performance than the six classical prediction indexes.Secondly,based on the node importance and path value,it considers the transmission capacity of the long paths among nodes,and the heterogeneity of paths is analyzed from the perspective of node importance.Then a Clustering Coefficient Significant Path(CCSP)index,a Betweenness Centrality Significant Path(BCSP)index and a Closeness Centrality Significant Path(CLSP)index have been proposed.The penalty coefficients of long paths in different networks are determined through experiments.Finally,experiments are carried out on 10 networks,and the improved indexes are compared with the classical indexes and the indexes in Chapter 3.The results show that the prediction accuracy of the improved indexes is greatly improved.And the CLSP index has good adaptive capability,which is of great significance to the development of the prediction algorithms. |
