Research On Characterizing And Modeling The Structure Of Complex Social Networks

21 century is the century of complexity and a major breakthrough in complexity research is predicted to be reality in the new century. As one of hopes to achieve this goal, complex networks attract increasing attentions from various fields. The focus in this dissertation is complex social networks which have been paid less attention than biology networks and information and technological networks. The reason for that is probably that due to the participations of intelligent agents it is hard to understand and explain the social networks. Two important themes related with the study of network structure are explored in the body of this dissertation. They include characterizing and modeling the structure of complex social networks. The contents of the former are built around the statistical analysis of real networks and concretely include quantitative methods of characterization of the structure and the analysis of accuracy and robustness. These three aspects respectively correspond to Chapter 2~4. From Chapter 5 to Chapter 8, the second theme is studied from the two angles of building and evaluating the network models. Concretely, Chapter 5 and 6 focus on the development and application of the research method of structural evolution of complex social networks. Chapter 7 and 8 focus on the development and application of the mechanism-inferring method that can be used in evaluating the models of complex social networks. In a word, the deep-going research on complex social networks will provide us the deep insight into the real social and economical complex systems and help us explore the common essence of various phenomenon and problems. In the sense of structure determining function, the focus on network structure in this dissertation is just the foundation of the deep-going research on complex social networks.The main contents in this dissertation can be summarized as follows:I. Study on a reference framework for the quantitative analysis of the structure of social networksIn the structural analysis of social networks, a new trend has shown that many attentions previously paid on the properties of single node or some kinds of nodes has been transferred to being on the statistical properties now. The contents of Chapter 2 are built around this new trend. We present a survey of the measurements used in the quantitative analysis of network structure both in traditional sociology and in complex networks. It includes the explanations and calculation methods of such measurements and the empirical results. This survey is hoped to present a clear and uniform reference framework that can be used in the structural analysis of social networks with different sizes in empirical and theoretical studies.II. Study on methods of estimating and testing for scaling exponents of power-law functions for complex networksIn Chapter 3, we explore the accuracy problem in the calculation of measurements: current graphical methods have the inaccurate faults in estimating the scaling exponents of power-law functions that are universal in the study of structural properties of complex networks. New methods are presented to improve the accuracy of estimation. Meanwhile, two statistics are introduced to test their performance. New methods are verified to be valid by application examples of CNN model-generated networks.III. Study on the effects of sampling on multiple structural properties of complex networksChapter 4 focuses on the problem of the robustness of network structural properties: whether the properties can be kept well when the network data is collected incompletely. We extend the current studies in two aspects including sampling methods and explored properties. Building the foundation of the analysis on a representative model of social networks with such multiple structural properties, we study the effects of different sampling methods on multiple properties. We qualitatively compare the changes of properties under sampling and, more importantly, quantitatively measure such changes by defining sampling distortion rates. Finally, a practicable way of implementing the hub sampling strategy is proposed.IV. Study on the development and application of the research method of structural evolution of complex social networksSince Chapter 5, we focus on a significant theme: network model research. From the angle of developing methodology, we explore the problem of modeling and analyzing the structural evolution of complex social networks. By reviewing the relative studies, a research method of structural evolution of complex social networks based on Dynamical Coupling Model of Network Structure and Agent Strategic Behavior (STC) is proposed. Due to the complexity in building and analyzing such kind of model, we present a four-element modeling framework and investigate the analysis methods of STC model. Especially, we provide a four-step process of application of regression technology to the model.Based on the above method, we build and analysis a concrete STC model in Chapter 6 to investigate the simultaneous evolution of network structure and agent strategic behaviors. Especially, we are interested in the evolution of structure affected by agent strategic actions. According to the four-element modeling framework, we extend the present studies in such four parts. Meanwhile, we introduce some quantitative measurements of evolutionary structure including degree heterogeneity, clustering coefficient and degree correlations. By applying the scenario simulation method and regression technology, we analyze the evolutionary results in three aspects: effect of dynamical structure on the emergence of cooperation in populations, effect of dynamical agent strategic behaviors on the emergence of structural properties and the quantitative effect of micro parameters on structural properties.V. Study on the development and application of the mechanism-inferring method for the evaluation of the models of complex social networksThe contents of Chapter 7 and Chapter 8 focus on another important problem in network model research that is the evaluation of models. In Chapter 7, we aim to develop the method of model evaluation for complex social networks. By analyzing the faults of current methods, we introduce a new algorithm of subgraph census and therefore proposed a quickly inferring method of network mechanisms based on the census of subgraph concentration. It is verified to be valid by applying it in a protein interaction network.In Chapter 8 we apply the above method to a representative of complex social network i.e. weighted scientific collaboration network. We review the current studies on models of weighted complex networks, particularly on models of scientific collaboration networks. After that, a classical data COND-MAT is adopted and two algorithms of edge-sampled RANDESU and node-sampled RANDESU are used to census the subgraph concentrations in this real network. We present the qualitative and quantitative analysis of mechanism-inferring results in four aspects including the analyses of prediction scores, accuracy and robustness, and the visual comparison of structural similarity between real network and model-generated networks. By these ways, we aim to provide the evaluation of prediction performance of various network models and propose the suggestions to improve current models.Summarily the distinct feature of this dissertation is the study of methodology for complex social networks. The concrete methods cover the significant parts of structure research including empirical study of real networks, structure analysis, network modeling and evaluation of models. Following are the primary innovativeness of this thesis in four aspects:â… . We have proposed new methods to estimate the scaling exponents of two important power-law functions. Compared with current graphical methods, new methods effectively improve the accuracy of estimation of two scaling exponents.â…¡. We have studied the effect of incomplete data collection on network structure and the strategy of solution. The original achievements include: (1) finding that sampling methods have a nontrivial influence on multiple structural properties of networks; (2) proposing a strategy of hub sampling based on the characteristics of real networks.â…¢. We have developed the research method of structural evolution of complex social networks and applied it successfully. This method solves the problem of how to model and analyze the structure evolution of complex social networks. Though the application of STC model in this thesis is not much deep, we provide a new way to do more research. The primary achievements includes: (1) demonstrating the key role of agent strategic behaviors played in the emergence of structural properties; (2) proposing the effective solution to the conundrum of collaboration for Hawk-Dove game from the angle of dynamical network; (3) providing the suggestions to adjust the micro factors to affect network structure based on the regression analysis for the STC model.â…£. We have developed the mechanism-inferring method for the evaluation of the models of complex social networks and applied it successfully in a weighted scientific collaboration network. Compared with current methods, new method can quickly infer the mechanisms of complex social networks that usually have large size and high density. In the application for the weighted scientific collaboration network, the evaluation result of prediction performance of various network models has been acquired and the suggestions to improve current models have been proposed.