Research On Identifying Influential Nodes Combined With High-order Structure Analysis

With the development of Internet technology,the way of communication between people has enriched,resulting in a large number of data.The network science research driven by data has become the focus of current research.Using the theory and method of complex network to model massive data is an important means of network science research.The traditional research based on the low-order connection mode of node and edge often ignores the ubiquitous interaction between nodes,which leads to the deviation of the whole network cognition.The current research shows that the network motif is an important functional module of the network,and the high-order structure can better describe the complex system.Accordingly,it has become a research hotspot to analyze and study complex network with network motif as basic research units so as to better regulate the network.Identifying influential nodes is a basic problem in the field of complex network research,which has important theoretical significance and application value.Most of the classical methods are based on the node and edge,which is low-order and micro-structure,and fail to further express the interaction behavior between nodes.However,the application of high-order network analysis to identify the influential nodes can not only reduce the dimension of large-scale networks,optimize the complexity of the algorithm,but also the interaction between individuals is considered.Therefore,the introduction of the high-order structure to identify influential nodes is more consistent with the structure and interaction characteristics of real networks,and the analysis results are more accurate and practical.The main contents of this thesis are as follows:(1)The high-order network analysis method is introduced into the directed unweighted network,and the adjacency matrix based on motif is obtained.At the same time,by reconstructing the network,a concept about high-order information is proposed to express the interaction between nodes.(2)In this thesis,evidence theory is introduced to fuse low-order and high-order nodes from different information sources in the network,and an algorithm of identifying influential nodes based on high-order network analysis is designed.At the same time,using semi-local centrality for reference to improve the proposed algorithm,a method of identifying influential nodes based on local high-order information is proposed.(3)The algorithm proposed in this thesis is verified by experiments on real datasets.We selected three different scale directed unweighted social network datasets.According to the existence forms of motif in specific fields,the specific number of different motif is counted.The motif with the most existence forms is selected to reconstruct the network and we get the corresponding high-order information.In order to better compare the difference between the proposed methods and the traditional methods,SIR model is used to analyze the propagation ability in the reconstructed network,and the robustness of the network is analyzed by removing the influential nodes in the initial network.The experimental results show that the influential nodes which combine the low-order and high-order information have better propagation ability and occupy a more important position in the network.