Research On Node Role And Group Evolution In Social Network

21 century is the century of complexity. As a powful tool for analyzing the complex science and complex system, complex network technology provides a brand new perspective. Much attention has been paid to the study of complex network by experts in different research fields for its wide range of applications. Hence, complex network analysis has become a hot research topic in recent years.A complex network is a set of items, which we will call vertices or sometimes nodes, with complex connections between them, called edges. In the recent 10 years, researchers studied deep inside the complex network structure with the beginning of the topological properties. They have proposed many centrality and distance metrics of nodes or edges, and have found that the networks often have the'scale-free' property, the 'small-world'phenomena, as well as the high'clustering-coefficients'.A social network is the graph of relationships and interactions within a group of social individuals. An online social network is the social network which extracted from the social media data. These two kinds of complex networks have been achieved a lot attentions, and studied extensively in the complex network science. In this paper, we discuss the node influence, node role in social networks, and the correlation of the social entities and topic evolution in online social networks. Analysis of node influence and node role using the social network analysis (SNA) method enables the discovery of the node's function in the network from a microscopic view. Finding the key nodes from the network may receive great practical value and realistic significance. And the analysis of the social media data using the SNA techniques could help us to find the interesting behavior patterns of the Web users, and provide technical support for innovative Web applications.The main contents in this dissertation can be summarized as follows:1.We propose a new information diffusion model CTMC-ICM, which introduces the theory of Continuous-Time Markov Chain (CTMC) into the classic Independent Cascade Model (ICM). The novel information diffusion model derived from the basic ICM such that a good estimate of?(A) can be efficiently computed, that is, the nodes influenced by a given set A of nodes.2. We give a new ranking metric named SpreadRank provided by the new information propagation model. SpreadRank is based on random walk theory, and takes into account the node's spreading ability. We introduce a different type of influence metric from the previous random walk based centrality measures by introducing the spreading ability into the transition matrix. We experimentally demonstrate the new ranking method which can in general extract nontrivial nodes as influential node set such that we could maximize the spread of information in social network, and it's more efficient than distance-based centrality.3. We introduce the structure properties of the network, and combines both of the node identity and the relational data for node role partition. The characteristic of our approach is different from the other behavior-based methods in that it adopts the social network analysis (SNA) methods. By taking into account of the network structure, the role defining process can benefit from two main respects:firstly, it takes account of the interaction data of the nodes which is very useful for understanding the node's value; secondly, it can give a global view instead of a local static view of the data. Finally, the methodology is applied to the telecommunication network. We identify the user role in the operators to help the telecommunication companies formulate strategies according to the users'different behaviors. 4. Two new social influence ranking metrics, InnerRank and OutterRank are proposed based on the concept of modified Pagerank, by considering the community structure knowledge. The InnerRank measures how well connected the node is to other nodes in the community, and the OutterRank is the measure of the links distributed among other communities. We hypothesize that the role of a node can be determined, to a great extent, by its InnerRank degree and OutterRrank degree, which defined how the node is positioned in its own community and with respect to other communities into 4 roles. It is well adapted to direct and weighted networks also. This method is shown to give reasonable results than previous metrics both on synthetic networks and real networks.5.Take online forums as an example, we analyze the online social media data on the Web. When the semantic links between the forum posts' content are formed, we build the temporal forum graph. In our analysis, the topic detection is in general a graph clustering problem. We cluster the temporal forum graph using the community detection method in social network analysis science, and obtain the results as the topics in the online forums.6.After the analysis of the topic detection on the forum data, we propose a method for discovering the dependency relationship between the topics of documents in adjacent time stamps based on mutual information measure. The knowledge of content semantic similarity and social interactions of authors and repliers are used to estimate the correlation between the topics. Furthermore, we define a new heterogeneous forum network structure, which include the information of both the semantic relations between the posts and also the publish-reply data between the authors. According to the affection degree of the social entities on the topic correlation evolution, we study the authors'impact and propose a new way for evaluating opinion leaders.Summarily the distinct feature of this dissertation is the study of methodology for social entities'position in complex social networks. Following are the primary innovativeness of this thesis in four aspects:1. We propose a new node influence metric, and extend the basic influence diffusion model. The new metric is able to improve the accuracy and efficiency effectively.2. Taking telecommunication network as an example, we adopt the social network analysis (SNA) methods into node role partition problem. This method is very helpful for understanding the node's value in the network, and gives a global view instead of a static view of the data.3.Two new social influence ranking metrics, InnerRank and OutterRank are proposed based on the concept of modified Pagerank, and by considering the community structure knowledge, to partition the node role into 4 roles. It can be well adapted to direct and weighted networks.4. We study the online forums represented for the online social networks, and proposed a new methodology to estimate the correlation between the topics using the knowledge of content semantic similarity and social interactions of authors and repliers. Furthermore, we study the authors' impact according to the affection degree of the social entities on the topic correlation evolution.