A Microblog Community Discovery Method Based On Intimacy Of User Behavior

In recent years, microblog has grown up as a new social platform in the Internet, which gradually aroused public concern. Under this background, the microblog network analysis has gradually become a hot issue in business community and academia. Through quantitative and effective data analysis on the virtual social networks, the information and rules hidden in the data can be revealed. In this topic, influence assessment of the nodes in the network and community detection is two classic questions. This paper makes certain improvements on the basis of existing algorithms on microblog network. The main work is as follows:In this paper, a new node influence evaluation model based on user behavior is proposed. This model investigates the users' forwarding, review, mention and private message and other behaviors, and constructs a framework to calculate the degree of intimacy between users. On this basis, the use of statistical regularity of forwarding and comments number makes the author put forward the concept of appeal power. Considering the user preferences of different types of information, we elicited the node influence calculation model ultimately. Compared with the traditional method which is based on Page Rank, the model takes more consideration of the information interaction between users. It abstracts out certain concepts from the related behaviors to carry on the quantitative research. So it is more aligned with the actual situation of the microblog network.In this paper, a community discovery algorithm based on node influence and intimacy distance is proposed. During merging central nodes, we take use of the edge weights for weighting and calculating node relevance respectively between nodes in fan set and concern set with the center nodes. In the process of the non-central nodes merging, the adjustment is carried out for the definition of network distance. The quality of the network connection is quantitatively investigated in the distance, which effectively weakens the effect of interference factors on the results. Thereby the non-central node partition algorithm is improved. This algorithm is applicable to find overlapping communities. Theory and experiments show that the algorithm has better performance in edge density, modularity and node's degree of community membership compared with Newman fast algorithm.