Cluster Structure Analysis And Dynamic Evolution Model Research Of Social Network

Social network is a virtual network describing real social activities, and the clusters of it embody the states of social gouphs which are naturelly formed by some reasons(such as family, colleague or same interests). Due to the social network is the portraiture of the realistic world, the features and development tendency of the real world canbe obtained through analyzing the structure of social network. Although analyzing the distribution of the nodes in communities and the correlations between communities can help to learn user characteristics and network topology, the study of social network structure cannot be confined to non-overlapping classification and clustering algorithms for the features and relations of nodes are varied and complex. Because different features of social network lead to different results of classifications, the users in social network can belong to several communities at the same time. On the other hand, the correlations between nodes are simple, so each edge canbe classified to a particular community. Therefore, supposing the only valuable knowledge of social network is the structure, the feature of the node is the combination of the features of the edges belonging to it. Social network allows users to generate personalized data independently, and the rich user information helps to analyze features of users in the round, but the data produced by users is poor standardization and uncontrollable, so the data is large and low quality. The difficulty of managing complex network information is increased, and the processed objects can not be limited to the writing knowledge, so the center nodes or expert nodes owning a large number of domain knowledge must be valued, which can extend the knowledge reserve of network. Besides, the attention paid on center nodes can help to learn the law of information dissemination, forecast development trend of network structure, and calculate the probability of changing status of nodes. For social network is a dynamic network, the stability of the analytical algorithms need to be controlled in a reasonable range, that is the over-emphasized stability leads to the insensitivity of new data, otherwise the wrong results gained due to the influence of the temporary information. In conclusion, the study of social network faces many difficulties. Whether from the aspects of data complexity or from the aspects of analyzing the dynamic network structure, a rigorous challenge is posed to the existing data mining algorithms, but the technologies from social search, personalized recommendation, to multiple positioning of users, knowledge map constructing are established on social network, which can not be achieved by any other past network platforms.To date, Internet gradually heads into full-social network age, and social network presents the features of micro-information and mobile data with the popularity of mobile terminals and the promotion of mobile network technology, which makes the network cover the life of people more comprehensive, therefore the study of social networks cause more and more attention from scientific researchers, enterprises, and governments. As Internet portals, such like Yahoo, are replaced by search engines, such like Google, social networks may replace search engines that exist a decade ago. In any case, social network with low operating costs and sticky services is changing the traditional pattern of Internet.This paper carrys on researches about social network from different aspacts. First, though the comprehensive analysis of the temporary features of nodes and the abilities of nodes to maintain inherent status, the community classification algorithm with strong stability is proposed. The algorithm does not merely depend on new data or existing features of nodes, but also uses the historical data of network topology, the abilities of nodes maintaining inherent states, and the extent of changes of new data, to calculate membership degrees between the changing nodes and the original clusters, compute the change degrees of nodes by combining the new data and the changing trend, and finally fullfil community classification and dynamic update. Second, an edge-bundling algorithm is proposed based on features, which illustrates the structures of complex networks clearly. The algorithm proposed outputs similar edges in nearby positions for clustering similar nodes and finding communities. Third, the algorithm proposed chooses the analytic target to be the center node, and calculates the probability of linking the center node and the undirectly connected node, which realizes the reasonable expansion of valuable knowledge and solves the small data problem of certain events for the non-overlapping classifying of edges is based on the features. The overlapping classifying of nodes can be achieved by analyzing and combining the node sets. Fourth, the information retrieval method of finding the users with a large amount of knowledge is proposed, which constructs the user interest distribution curve and calculates the slope of the critical point for finding expert nodes. Ultimately, the experts can be found from user base, and the experts in this paper contain not only authoritative ones, but also normal ones. Fifth, the knowledge representation method and architecture fusion strategy is proposed by mining common structures of multi-texts to find implicit sematic relationships and increase the accuracy of text matching. The algorithm can analyze many texts in the meantime and only scan a text one time in the process of text analysis, which greatly reduces the computational complexity and improves the accuracy of text matching. Sixth, using Bayesian network to reduce the complexity of the probability model, and analyzing influences between nodes in low density network to predict the status of the center node and learn the change tendency of the network structure, the mehod proposed calculates mutual effects between nodes to analyze the status of the adjacent nodes and predict the movement trend of the center node according to three different relationships and shortest distances between nodes. Finally, the experiments in different chapters compare the algorithms proposed with the other similar algorithms using different datasets, and output the comparison results visually, which verify the feasibility and correctness of the algorithms.