Research On Node Importance And Community Detection Based On Coulomb Force Centrality

Complex network is an abstract model for understanding complex systems in the real world.It represents the entity objects in the complex system as nodes,and the ex-plicit or implicit relations between entity objects as edges.As two of the most popular topics in complex networks,the identification of critical nodes and the detection of com-munity structure have been widely studied and applied in many fields such as computer,communication,criminal investigation,society,finance,transportation,biology and so on.Critical nodes are the core elements that constitute a network and realize its informa-tion transmission function.Identifying and protecting the critical nodes can effectively improve the anti-destruction of the network,and can also propose more efficient network attack strategies starting from the critical nodes.The community structure in a complex network is a special property,in which the connections between the nodes within the community are relatively close,while the connections between the nodes are relatively sparse,and the process of discovering the community structure in the network is called community detection.Community detection can easily reveal the hidden structure of the network,reflect the relationship between nodes,analyze the potential information of the complex network and predict the local functional characteristics of the network.This paper studies the node importance assessment method and community structure detection method in complex network,and the main results are as follows:The k-shell is a common measure of node importance.But due to its shortcomings such as destroying the overall structure information of the network and ignoring the in-fluence of neighboring nodes,it is difficult to ensure that each node can be quantitatively distinguished.In order to improve the accuracy of node identification,this paper firstly improves the decomposition process of k-shell and proposes the Ak(Accurate k-shell).Considering the influence of local feature information and global structure information on nodes in the network,then the Ak is applied to the GC(Gravity Centrality)and pro-pose the CFC(Coulomb Force Centrality).Because Shannon entropy in informatics has good expansibility in identifying critical nodes of the network,through the Shannon en-tropy of combination neighborhood degree centrality,neighborhood accurate k-shell and coulomb force centrality,the MC(Mixed Centrality)is finally proposed.This measure can fully integrate the topology information,location information,local feature informa-tion and the overall topology information of the network.In addition,it can eliminate the one-sidedness when a single measure is used to evaluate nodes,so as to achieve the purpose of evaluating the importance of nodes pluralistically.Under 7 kinds of real net-works,conducting a series of experiments on the monotonicity and accuracy of MC and other measures,the experimental results show that MC has better performance in identifying critical nodes.Label Propagation Algorithm(LPA)is an efficient and fast community detection algorithm,which can be applied to networks of different sizes.However,due to the ran-dom propagation of labels and the ambiguous convergence conditions,the experimental results are very unstable.In order to further identify the critical nodes in the network,this paper firstly proposes the Union Centrality UC on the basis of CFC.Then,for solving the defect of randomly selected labels in LPA,this paper uses UC to design LS(Label Selectivity),which is used to evaluate the propagation ability of node labels.is proposed to evaluate the propagation ability of node labels.By applying UC and LS to LPA,this paper finally proposes a label propagation algorithm based on coulomb force centrality(CFCLPA).This algorithm can eliminate the randomness in LPA and define the ending condition of the algorithm.Therefore,it has the ability to identify the com-munity structure efficiently and accurately.A series of tests and comparisons are made between the real network and the LFR benchmark network,and the experimental results show that CFC LPA has better community detection performance.