Research On Microblog Community Detection Via Core Tags

With the rapid development of the Internet, various kinds of community network emerge in the Web. The community network can represent its social activities. While the network is organized into a form of community, it not only make people understand the organizational form of resources and development trend of information but also serve as an effective organization of the hierarchy for the network through identifying and distinguishing the community. So currently the community detection has become a research hotspot.As a social new media, microblog has accumulated hundreds of millions of users in just a few years due to its short, real-time and openness and its number is still growing rapidly now. As it brings the increasingly widespread social impact, community detection for microblog user demonstrates its important theoretical and practical significance. The traditional microblog community detection algorithm has the characteristic of low coupled clustering and the overlapping degree can not be controlled. In this paper, we present a divisive approach for overlapping microblog community detection algorithm via core tags. The main contribution of this thesis is organized as follows:1) The co-occurrence between tags and inverse user frequency is considered during the tag expansion stage. The tag weight for each user is decided based on both the inner and outer correlation of tags.2) Three evaluation criteria are defined to form an evaluation function, which focus on user attribute and following information. Concretely, cohesion, coupling degree, and practicality are defined to modify the division of community, and then an iterative mechanism is designed to obtain the optimal result.3) A divisive approach for overlapping microblog community detection algorithm is presented based on core tags. Whether an independent community will be constructed or not depends on the overlap user number with the existing community, which thus achieve the overlap degree. Two different algorithms are proposed based on two terminating condition.