Research On Techniques Of Target-Node Analysis In Complex Networks

The research of complex network is a fusion interdisciplinary of topology, graph theory, social network analysis, Internet macroscopic topology, physics, biology and other fields. The methods of this discipline consist its own hierarchical structure. Members of a small football club, or the boundless cosmos, they can be considered by the same approach, which most scientists in the field believe, "Structural and dynamical similarities of different real networks suggest that some universal laws might accurately describe the dynamics of these networks." Meanwhile, control theory, game theory, queuing theory, hyperbolic geometry and other natural science theories have been found composable with complex network research. While people found general characteristics and laws of complex network within different real-world complex systems, the research of complex network is proved useful in many disciplines. However, existed models and analysis methods still can’t reconstruct or explain a few features found in real complex networks.On the other hand, the research on techniques of target-node analysis in complex networks is based on complex network theory and focuses on selected nodes in networks, using centrality measurements, community discovery, link prediction, statistics, etc., to explore and analyze target network. Based on the understanding and characteristics found in the network, related research usually presents novel methods and extended model, design test process and confirm the effectiveness of proposed methods. For a general complex network research, the process of the research can be divided into three stages,1) detection or abstraction of the network,2) exploration and analysis of the network, and 3) design a novel or extended model to solve a problem or demonstrate a phenomenon.For the three stages of such a research process, this dissertation includes four aspects to explore research techniques of target-node analysis. They are as follows. (a) Design a novel detection algorithmn of complex network, (b) propose a measurement to evaluate the importance of nodes in networks, (c) classify the evolution of complex network into normal and unusual types, (d) devise a novel model to solve a given problem.First, a detection algorithm is designed to abstract complex network from Chinese text. By weighting frequency of edges based on their topology characteristics, detection of lexical network within the text is simplified. With low time complexity and various applicable targets, experiments prove that the abstracted complex network structure is with standard small-world features and approximate scale-free characteristics, which confirms the validity of proposed algorithm.Second, to improve measures weighting importance of nodes in complex network topol-ogy, a novel centrality metric called neighbor vector centrality is presented. In attack simula-tions on various real-world networks and synthetic networks, our result confirms the effective-ness and generality of the proposed centrality. In most of the simulation test, neighbor vector centrality outperforms other three standard centrality measurements.Third, based on observation of a real-life complex network evolution(Internet IPv6 topol-ogy), unusual evolution of Internet topology is detected and defined. IPv6 Internet topology evolution in IP-level graph is analyzed to demonstrate how it changes in uncommon ways to restructure the Internet. Two kinds of unusual evolution are selected and explored. After eval-uating the changes of average degree, average path length and some other metrics over time, it confirms that in the case of a large-scale growing the Internet becomes more robust; whereas in a top-bottom connection enhancement the Internet maintains its efficiency with links largely decreased.Finally, based on a recent important work published in Nature, a Resource-Competitor double-layer model is proposed (RC model). Using the RC model, founding process of nine dynasties in Chinese history is analyzed. It shows that competitor with more drive nodes has more influence on the control of the network. Therefore, such a competitor has a greater pos-sibility to win the control power over the whole system. Using data collected in January 2012, a prediction result on members of political bureau of the central committee of the communist party of China is given. The real result in October 2012 confirms 81% prediction accuracy.In a word, our work concerns novel measurements and techniques for complex network analysis, and mainly discusses how to analyze selected target nodes in the network, includ-ing node(network) abstraction, node importance measurement, classification of node(network) evolution and the design of a novel model. The validity of those approaches is confirmed by analysis or experiments to make sure that they are simple, easy to use and effective.The given methods are with various applicable target networks. Our research proves that they are necessary for theory and application.