Research On Clustering Method In Complex Networks And The Application In Analysis Of Topics Of Public Opinion

In big data era, large user groups and the connection among users make that the social networks have become a major medium and key points in the internet public opinion. Micro-blogging has become the media representatives of Web2.0 and a source of the information of the internet public opinion. Thousands of messages are generated by users through Micro-blogging. The research on the networks of users and texts are needed in order to conduct comprehensive Micro-blogging network analysis. It is crucial for internet public opinion to discover overlapping communities from social networks effectively and accurately identify different topics from Micro-blogging articles.On the basis of the clustering algorithm of complex networks, this dissertation improves the original fuzzy clustering algorithm and realize the discovering overlapping communities from complex networks Researching the text clustering in particular to analyze the text topic extraction so that discover accurately the information of internet public opinion and master the dynamics of internet public opinion. The main contents of this dissertation are as follows:(1) This thesis introduces the research status of clustering algorithm of complex networks and topic analysis of internet public opinion. At the same time, the basic concept of complex networks, community structure and the measure standard of partition of community structure are described in detail. The main algorithms of discovering community structure in the complex network and the algorithms of discovering overlapping community is described in detail and is analyzed(2) The relation between clustering algorithm and discovering community structure of networks is proposed based on clustering theory. Q value, the measure quality indicator of discovering community structure, is introduced on the basis of fuzzy clustering algorithm and solve the difficult issues of determining the number of clusters and discover the overlapping community structure in complex networks based on improved FCM algorithm. Using improved FCM algorithm with both classical datasets, it is proved that the improved FCM algorithm can effectively detect overlapping community in complex networks.(3) Based on the introduction of the topic clustering model in detail, The Dirichlet allocation model with Gibbs sampling is used for analysis on the text topic. The document-topic and topic-word probability matrix is obtained to infer the implicit text topic variable. Using LDA model analyze the data of World forum to achieve the text topic mining. Finally, the results of experiment are analyzed.(4) Using clustering algorithm conduct the analysis of top of text in accordance with the microblogging network. The microblogging network is divided into two levels including the user level and blog text level. With real Sina blog data, detecting the overlapping community based on improved FCM algorithm in the social networks formed by users and finding the text topic in blog level formed by microblogging using LDA model. Analyze the user-blog networks based on method of detecting overlapping community. The experimental results show that the distribution of blog published by users in overlapping community is power-law distribution. The results also show that community size is positive correlated with user and there is overlap between the topics of micro-blogging published by users from the same community.