| With the rapid development of information technology,the human society has entered the network era since the 21 st century.It can be said that our lives are surrounded by various networks.In this paper,aiming at the defects of the research on link prediction in complex network,guided by the theory of the complex network,link prediction and rank aggregation,we introduce rank aggregation method to link prediction in complex network to aggregate the list of unknown link possibility.Around the three problems of characterization of the links collection in complex network with incomplete information,the definition of energy matrix in rank aggregation and the choice of link prediction algorithms based on the similarity of network structure,we systematically studied the problem of link prediction algorithms selection in complex network by integrating multidisciplinary of the statistical physics,the traditional knowledge of graph theory and matrix theory,probability and statistics,control simulation,and put forward and discussed the link prediction algorithm based on the aggregate.At the same time,we analyze and discuss the selection of the link prediction algorithms in complex network.This paper’s main research and innovation points are as follows:(1)Put forward the new algorithm of link prediction based on rank aggregation.Researching and analysing the link prediction from the perspective of complex network,we find that the link prediction algorithm is to predict the existing possibility of unknown links,and select the unknown links with high existing possibility as the hidden link information in incomplete complex network.Based on the essence of link prediction algorithm,we introduce rank aggregation method into the link prediction of complex network,and propose the link prediction algorithm based on rank aggregate.Based on the existing link prediction algorithms which are depending on network structure similarity,we aggregate the existing possibility of unknown link list by the rank aggregation,and get the new list of unknown list.Furthermore,we can predict the missing links in incomplete complex network.The study found that the accuracy and efficiency of link prediction algorithm based on rank aggregation can improve a number of times compared with adding edges randomly.(2)Put forward the new way to the choice of link prediction algorithm which is based on the algorithm which we put forward before.In the real application,considering the role of the complex network link prediction algorithm based on rank aggregation is so "smooth",and the precision of the algorithm can only in the medium level,so it is difficult to apply in the cases which require higher precision.In this paper,we regard this algorithm as the "reference answer" in the link prediction under the condition of incomplete information,using predicted networks obtained by other link prediction algorithms compared with the "reference answer".By computing the spearman simple weighted measure between the two values,we choose link prediction algorithm with the minimum value measure as the most suitable algorithm for the specific network.Further,if the two or more algorithms have the same measure values,we compare the AUC value of the selection algorithms,and select the link prediction algorithm with the highest AUC value as the suitable algorithm for a specific network.Studies have shown that the way of link prediction algorithm choice based on rank aggregation method can effectively choose the suitable algorithm for the network under incomplete information. |
