The Kernel Framework Based On Latent Features For Link Prediction And Networks Disintegration

Link prediction in complex networks is prevailing in physical and computer science,for the algorithm can be used to extract missing information,identify false interactive information,evaluate network evolution mechanism and so on.Under the premise that data access to information in the biological hacking group is not complete,biological hacking group can be destroyed by introducing link with data mining as well as the evolution of network reconstruction effect to the collapsed model of complex networks.In this context,a nuclear framework for network reconstruction and link prediction is proposed,and the network disintegration analysis is carried out in the data of biohacker organizations.The detailed work is as follows:Firstly,a kernel framework based on non-negative matrix is proposed to solve the link prediction and network reconstruction.The proposed kernel framework takes full account of the links in between the network nodes,and can better learn about potential characteristics of the network through kernel mapping.By introducing the kernel function,the local information of the network is combined with the global information to make the similarity matrix denser and thus better revealing the network structure.Secondly,the kernel matrix factorization algorithm is proposed.The proposed novel method of dealing with the similarity matrix is to combine the matrix factorization and kernel function.Through using the kernel function,two matrix factorization matrix factors are embedded into the Hilbert eigenspace of high dimension.At last,the similarity matrix of nonlinear reconstruction matrix of the original space is utilized to predict the possibility of the connection in between nodes.Finally,we use the constructed model to analyze and verify the real biohacker data.Based on biosecurity,the target network of biohacker groups is applied.By drawing the network structure diagram of the biohacker group,the author makes a simulation analysis of the link prediction and disintegration effect of the proposed algorithm on the target network.The results show that the two proposed algorithms presented in this dissertation are prominent in the empirical network.