Research On Community Detection And Visualization Based On Micro-blogs Interaction

With the development of Internet technology and the rise of social networks, social media has become the mainstream in the Web2.0 era gradually replace the traditional media. Social media is a new kind of internet online tool and platform that people can spontaneously disseminate information, share experiences and express their own emotions. Among them, Micro-blogging is the most typical one that it has become an advanced research hotspot in the social network visualization analysis area. Social network visualization analysis focuses on the relationship mining algorithms and visualization analysis. Traditional relationship mining algorithms can be used to detective different communities, but most of them will increase the computational complexity to some extent in the process of overlapping communities. Especially, the performance of the algorithms will drop quickly in the large-scale network structure. As a result, the label propagation algorithm emerges at the right moment to solve this problem. But its randomness in the stage of label initialization, label transmission and label selection has a bad influence to the stability and accuracy of the community detective result. Visual representation of social networks with visualization technologies contributes to social network analysis. Therefore, it has a certain experimental value as well as application significance to establish a network visualization analysis platform.The self-forming social relationship networks of social media are a reflection of the social relationship networks in real life. In this paper, it mainly analyses the advantages and disadvantages of SLPA label propagation algorithm and put forward a new improved algorithm and then design and implement a network visualization analysis platform. The main works of this dissertation are;(1)Introduces the research status of label propagation algorithm on disjoint community detection as well as overlapping community detection, and describes the core idea of SLPA algorithm, analysis the instability of community identification result from the randomness in the label selection and transmission phase. And for this reason, this thesis presents a collaborative filtering Speaker-Listener label propagation algorithm, which has been optimized labels initialization and used collaborative filtering algorithm to update a label in the label selection phase. To inspect the performance of the improved algorithm, this thesis presents experimental analysis between synthetic and real networks. The experimental results show that CF-SLPA algorithm has better accuracy and stability.(2)Integrate theory with practice to design and implement a social network visualization analysis platform that integrated with data collector, data analysis, social network analysis, community detection and application.(3)System demonstrations show an example of its usage and its important practical application value. The results show that the platform is benefit to the research, and it also offers a valuable experimental application platform to other social computing.(4)Finally, summarizes the content and points out the further study emphases and direction.