| In recent years,network science has become one of the hot topics that researchers in various fields pay wide attention into it.The study of networks is of great significance to the understanding of real systems.Scientists have successfully applied the theory of networks to many real world systems,such as neural networks,power grids,protein interaction networks and social networks.There are various types of meso-scale structures in networks,among which the two most widely studied meso-scale structures are the community structure and the core-periphery structure.There are a lot of regular network structure features hidden in the meso-scale structures of networks.Revealing and understanding meso-scale structures in these networks provides important reference information for further understanding the functions and characteristics of networks.In this thesis,we will focus on the key concept of time series,and construct the time series model in the networks to complete the analysis and detection of the meso-scale structures in the networks.Our work mainly includes the following three aspects:(1)Detect the community structure in the network based on time series.According to the pure structural characteristics of the network,the deterministic parameter-free diffusion model in undirected network,the deterministic parameter-free signal copying model in directed network,and the deterministic parameter-free signal copying model are respectively established in dynamic network to drive the flow of information between nodes in the network,thus generating the corresponding time series.Equipped with the time series,the corresponding dissimilarity index functions are constructed for the different types of networks to describe the similarity between nodes in the network.On the basis of the dissimilarity index,we design the community detection algorithms in the undirected network,directed network,and dynamic network,respectively.(2)Detect the core-periphery structure in the network based on time series.Based on the pure structural characteristics of the network,a deterministic parameter-free signal copying model is established in undirected network to describe the dynamic evolution of node signals in the network,and obtain the corresponding time series.The time series records the amount of signals received by each node in the dynamic process.On the basis of the time series,a new centrality measure called time-series core score is constructed to evaluate the importance and influence of nodes in the network.Therefore,we apply the proposed centrality measure to detect the core-periphery structure in the synthetic networks,and propose a separation algorithm to detect the core-periphery structure in the real networks without prior division.(3)Fusing the dynamics and network topology information to detect the community structure in the network.We consider a fusion strategy,which utilizes both the dynamic behavior of cascading failure and the local topology characteristics of the network to fuse the global dynamic information with the local topology information.We propose a new hybrid clustering community detection algorithm,which is divided into two stages.In the first stage,the similarity measure between nodes is calculated based on the dynamics information in the framework of the k th-nearest neighbor density to guide the merging of nodes in the network.In the second stage,the local topology information is employed to construct the corresponding metric function to guide the further merging process of the remaining node groups. |
