Community Detection For Microblog Based On Link Analysis And User Interests

As a new social media, microblog has been developed rapidly in recent years and become a significant communication platform gradually. Microblog users can make up variety of virtual communities. It is important to detect users' communities with highly common interests and denser network structure effectively, and the result can help us to improve the accuracy of self-oriented services, target marketing and link prediction.However, the traditional community partition methods usually lack of the comprehensive thought that referring to the link relationships between nodes and the nodes' contents, which leads to lower cohesion in interests of communities divided by these algorithms. Some methods that base on link relationships between nodes and the nodes' contents use the microblog contents to analyze the users' interests, but there is lots of noise in the users' microblog contents and the microblog is often issued randomly, due to those factors, the content of the microblog may not reflect the users' interests accurately. According to this background, we have studied on the users' profile and find that it can reflect the user's characteristics. In this thesis, we call the information that can well reflect the users' characteristics as users' interest information.In this thesis, we study on the problem of microblog community discovery based to the characteristics of the microblog network after having analyzed and summarized the existed community discovery methods. Then we present a micro-community detection method based on the users' links and interests. The community with denser common interests and network structure that divided by our method can better reflect the real social relations. The main contents are as follows:Firstly, we study the traditional community discovery methods and summarize the advantages and disadvantages of these algorithms. The existing community evaluation methods are classified and introduced in this part.We also make a simple description of the network analysis tool.Secondly, this thesis introduces two famous microblog platforms and makes a simple description of the characteristics of the microblog network. Then we introduce the shortcomings of the two frequently used community detection methods. The community discovery approach that based on user interest is introduced.Finally, this thesis analyzes the link relationships between microblog users and users' interests, then we deduce the formula of link similarity and user interest similarity, and combines these two similarity measures to calculate the total similarity between users.According to the total similarity between users, microblog network is transformed into a undirected weighted network. Based on the Louvain algorithm, we propose a new method todiscover the community in the microblog network. We get the community visual results by using the social network analysis tool Pajek to obtain a more intuitive understanding of the community. At last, we verify the effectiveness of our method by comparing it to the existed community detection methods for some classical evaluation criterion.