The Approach To Discovering Multiplex Structures From Complex Networks

In the analysis of complex network, structure analysis is a very important subjectin that it’s the basis of correctly understanding network function and analyzingnetwork dynamic behavior. In the past few decades, a series of structures are foundand analyzed such as community, outlier, hub, authority, multipartite, bow-tie, etc.Recent studies show that there might be many structures coexistent in one network.These structures are in a complex relationship of overlapping, interaction, and nesting.The structures and the relationship among them together form a heterogeneoushierarchical structure called multiplex structures, which provides a new perspective ofunderstanding the influence of network structure toward the network dynamic andfunction. Bo Yang etc. have done a deep research on the multiplex structures ofnetwork, and further propose a granular stochastic blockmodel (gSBM) whichcharacterizes multiplex structures and the corresponding approach to discovermultiplex structures. However, since this algorithm is very complex in timedimension, it can’t be widely used in analyzing large-scale real networks which arealways the case for real-world networks. To effectively decrease the complexity intime dimension of existing multiplex structures’ discover methods, this paperoptimizes the proposed gSBM and its learning method, based on which, furtherdevelops a multiplex structural prototype system for discovering structures and visualanalysis. The main work done in this paper is as follows:(1) From model selection perspective, optimize the model selection strategy of theoriginal gSBM from serial to parallel by adopting competitive learning modelselection, so that model selection and parameter evaluation could be combined intoone iterative estimation process. By this way, learning time of the algorithm would beoverall decreased and the algorithm efficiency would be significantly improved.(2) Develop a corresponding prototype system which has the following functions:process various formats of dataset files in complex network, compute commonstatistical properties of complex network, mine multiplex structures in given networks,display graphically the multiplex network structure in visualized GUI, and generatesupporting files for deeper analysis. (3) Verify the effectiveness of the proposed approach and prototype system bydoing experiments in both artificial synthetic networks and real networks. Throughbenchmark datasets, this paper analyzes and verifies the proposed approach, andcompares the approach with typical ones in computing quality and performance. Itturns out that the approach proposed in this paper is better than others in timecomplexity. In the real network experiments, multiplex structures samples in reallynetworks are illustrated, introduced and analyzed from multiplex structuresperspective, and multiplex structures are verified to be widely existed in real networksby empirical analysis.