Mining Relationship Between The Characters Based On Complex Network Analysis

In the real world,complex systems can often be abstracted into network topologies consisting of nodes and edges.With the in-depth research on complex systems,there are two problems in complex network analysis methods.Firstly,the complex network model is developing towards heterogeneity and diversification.The traditional complex network topology is a highly abstract description of complex systems.With the in-depth study,the heterogeneity of complex networks is becoming more and more important in the research of network.It is a vital research direction that how to analyze the algorithmes of heterogeneous complex networks.Secondly,the network size is growing.The surge in the amount of data poses a serious challenge to the storage and computation of complex network algorithms.It is an important problem to extract an approximate thin structure from a large-scale network topology.In order to solve the above problems,this thesis is based on the edge pattern to study the complex network.The existing research results show that the edge pattern can help to study the network properties of node attribute relations,the network generation model and the high order topology expression and so on.In this thesis,the heterogeneity of complex networks and the influence of core structure is studied by using the research method of edge pattern combined with the traditional complex network analysis theory.The main contributions of the thesis as follows:1.Proposing an algorithm for overlapping community mining based on the multilayer network model.The thesis systematically studies the link community detection(LCD)algorithm,which is an efficient overlapping community mining algorithm based on link-pairs in single layer network.Besides,an improved algorithm,which can be used in heterogeneous relationships of complex network models,is proposed based on the flaw of the original LCD,and the applicability of the algorithm in the multi-layer network model is analyzed.Finally,the LFR framework,which is a community performance detection algorithm,is utilized to compare the results of the proposed algorithm with the Louvain and Infomap algorithm,and the validity of the algorithm is confirmed by the result.2.Proposing a complex network core structure extraction algorithm.The algorithm excavates the core model instances in neighborhood subgraph of each node in the network,and then merges them into the core impact structure.Different from the traditional core structure mining method,the core impact structure is a reductive network subgraph,which not only contains the core structure of the network,but also describes the influence of core structure on non-core structure which is of great significance for the analysis of network communication and influence modeling.Finally,it is found that the core influence structure can reflect the topological characteristics and scale characteristics of the original network by experiment.Meanwhile,because the core impact structure retains most of the characteristics of the original network,and a good streamline network structure,the method can also be used in the field of network visualization.In summary,the thesis makes a deep research on the relationship heterogeneity and the core influence of complex network based on the link pattern as the basic unit,and has achieved remarkable results.Edge pattern,which is a more complex network basic unit,can be used as the future direction of complex network science research.