Research On Network Inference Method Based On Network Node Complexity

With the development of network information technology,a large number of complex systems appear in large numbers.Meanwhile,the research on complex networks also develops rapidly.During the study,the data in the network structure was incomplete.One is that any structure of the Network is not clear,and links between nodes can be inferred based on the attributes of nodes in the network structure.This kind of problem is called Network Reconstruction.In another case,a part of the Network structure is unknown,and even the node time series information of this part is unknown.This kind of problem is called Network Completion.This paper studies the above problems from the similarity and importance of nodes.The main innovative work of this paper includes the following two parts:(1)A network reconstruction method based on node Similarity(SDS)is proposed.The network structure is reconstructed according to the similarity of nodes in the network.In short time series,the stronger the coupling relationship between nodes is,the change trend of corresponding nodes is the same.The consistency level of the transformation trend among nodes was quantified by the sorting method in SDS.The higher the consistency level of the change trend of nodes,the stronger the coupling relationship between nodes and the existence of links between nodes.The SDS method was compared with Kendall Rank correlation coefficient,and the influence of noise and number of nodes on the experiment was considered.SDS method takes into account the ordering information ignored by Kendall Rank correlation coefficient and has low time complexity.Therefore,in a large number of simulation experiments,SDS method is superior to Kendall Rank correlation coefficient in the accuracy of network reconstruction,which indicates that SDS method is suitable for short time series.And it is effective in network reconstruction.(2)In this paper,an Excess Average Degree Common Neighbors(EADCN)method based on the importance of nodes is proposed to complete the network according to the importance of nodes in the network.The more neighbors two nodes have in common,the more similar they are and the more likely they are to be linked.However,in the common neighbors of two nodes,and non-important nodes.Whether there is a link between two nodes should not only be measured according to the number of nodes in the common neighborhood,but also the importance of nodes in the common neighborhood.The EADCN method adds the Excess Average Degree to the common neighbor,and uses the excess average degree to calculate the average degree of the neighbor nodes of the nodes in the common neighbor.If the value of the residual average of a node in the common neighbor is large,it indicates that the node is important.Because the EADCN method fully considers the importance of nodes in the network structure,the accuracy of the EADCN method in network completion is generally better than that of the common neighbor(CN),Adamic-Adar(AA)and Resource allocation index(RA)methods in a large number of comparison experiments.This shows that EADCN method is effective in network completion.