Research And Implementation Of Cluster Analysis And Visualization For Social Network

Social network includes the interrelationship between entities.We can extract the network's composition pattern and evolution law by analyzing these connections.However,social networks are often large-scale,complex in structure and dynamic over time.Therefore,how to analyze social networks and extract effective information from them is a meaningful and challenging research topic.Though clustering can effectively solve the problem of large-scale network data,there are some problems in the process of clustering,e.g.,it is difficult to determine the number of clusters and the clustering results are unstable.In order to make the cluster number as accurate as possible,in this thesis we propose a hierarchical visualization aided method based on the existing clustering algorithm DCN(Community Detection based on Centers and Neighbors),and improves the label propagation algorithm to solve the problem of missing cluster and unstable clustering results.In addition,large-scale networks have some special structures compared with toy network datasets for experimental analysis.Since large-scale networks are often not strongly connected graphs,there will be a large number of isolated structures with close internal links,and the number of such structural nodes is small,which makes them easily ignored in traditional clustering analysis.Therefore,in the process of clustering,not only large connected sub-graphs need to be analyzed,but also these isolated clusters need to be detected,so that the results of clustering can represent the overall distribution of the network.In order to enable users to participate in clustering analysis and dynamic network exploration,in this thesis we design a visualization method of social network,and realizes a prototype system based on Web combined with clustering analysis and exploration.We use the actor collaboration network extracted from Douban Film Data to carry out case analysis,which verifies the practicability of the system.The main work of this thesis is as follows:(1)Visualization-assisted clustering algorithm.In order to solve the problem of missing cluster center points and unstable label propagation results in existing clustering algorithm DCN,we proposes a visualization-assisted method to enable users to identify potential community center points and participate in the clustering process.Besides,the importance-based label propagation algorithm is proposed to make stable the label progression process.In order to reduce the difficulty of detecting potential centers in the network with much clusters,this paper combines the hierarchical clustering idea and enable users to identify potential centers iteratively.(2)Clustering analysis of large-scale network data.In order to detect the large number of isolated structures in the large-scale network,we identify the isolated structures by analyzing the first-order neighborhood of the nodes with maximum density.In order to detect local centers in large-scale networks,we regard nodes that cannot assign labels as centers in label propagation process,and re-propagation labels based on the updated list of centers.(3)Visual analysis for static and dynamic social networks.In order to show users the static structure and dynamic evolution of the network,we design visual views for displaying network structures of different levels.We design and implement a prototype system to enable users to participate in clustering analysis and visual exploratory of the network.