Study On Social Network Analysis Based On Topic Model

Social network analysis is the analysis of the social network graph, and the ad-ditional information within the graph. In recent years, with the rise of online social networks, the social network analysis has brought new challenges and opportunities. Link prediction and community discovery are two very important problems of social network analysis. Link prediction is task of predicting the link that is missing link or being arise in the future. Community discovery is to find a few sets of nodes, where each node set contains some similar nodes in the social network. In this thesis, we focus on the study of Social Network Analysis based on Topic Model with node text information and structural information.Firstly, the social network analysis is described in detail, a number of research in social network analysis, especially its two important research:the classical algorithms of link prediction and community discovery.Secondly,a topic model-based collaborative evolutionary link prediction algorith-m is presented. The algorithm firstly use LDA model modeling text data; update the the document topic distribution parameter using of the idea of collaboration and evolution; calculate LDA-based document similarity matrix; build a new network structure using the similarity matrix and the adjacency matrix; calculate resource allocation(RA) mea-sure based on the evolution of collaboration; run the link prediction task. Experimental results validate the method and show that it is more effective in link prediction task in the case of that network contains few links and text information attached to the node.Finally,community discovery algorithm based on the topic and the idea of collab-oration and evolution is proposed. The algorithm firstly use LDA model modeling text data; update the the document topic distribution parameter using of the idea of col-laboration and evolution; calculate LDA-based document similarity matrix; calculate resource allocation(RA) measure based on the adjacency matrix; fuse the two similar matrix into a new similarity matrix; run the community discovery task with spectral clustering algorithm based on the new similarity matrix.