Research On Key Issues Of Analyzing And Mining Online Social Networks

It is a challenging research direction for social networks for its complex structure, different behaviors of users and its huge amount of data generated by various events of all users, while it attracts attention of many researchers from various research fields for such reasons, and substantial achievements are made by them. However, as a new emerging thing, online social networks is still developing, there must be a lot of problems waited to be studied.In this dissertation, several key issues are discussed and researched on the distinguishing features of individuals groups and characteristics of the network structure. The main research contents are as follows:(1) On the aspect of the individual analysis of social networks:This dissertation focuses on calculation of the relationship between users and topics based on the contents published by the users. Base on the contents published by the users themselves and exploiting the non-negative matrix factorization technique, some user-topic sensitivity calculation algorithms are proposed. These algorithms can analyze the contents published by the users and compute the relationship between the users and topics effectively. The experimental results on real world datasets show that the proposed algorithm can effectively analyze users'published content, and calculate the user-topic sensitivity.Specially, as the prevailing of microblogging, the situation of information overload is getting worse, and more and more users tend to post short text data. Short text made the data very sparse, and this dissertation proposed a user-topic sensitivity algorithm based on the co-occurrence of word. Ordinary algorithm can not handle such short text sparse data well, but the proposed algorithm is particularly designed for short text, which can effectively avoid data sparseness problem and effectively calculate the results. Experimental results on real world datasets show that the algorithm proposed in this dissertation can avoid the sparseness problems and calculate the results effectively.(2) On the aspect of the network structure mining:This dissertation puts forward concepts called critical nodes and critical blocks, and designs an effective and efficient algorithm to find such special nodes in the network. Different nodes have different importance in social networks is acknowledged by people, to measure the importance of nodes many measurements and algorithms have been proposed, such as all kinds of centrality, concepts of k-shell and k-core, and a variety of algorithms based on PageRank and HITS. In this dissertation, some types of nodes which are called critical nodes are believed important in social networks from a different view. By using the properties of the Fiedler Vector of matrix, a heuristic algorithm is proposed to find critical nodes and critical blocks. Experimental results on real world datasets show that there do exist critical nodes, and the proposed algorithm can find them effectively and efficiently.(3) On the aspect of group analysis of social networks:This dissertation studies the community discovery problem. The problem of community discovery has attracted many researchers, however, most of the published studies are based on the structure of the network. Considering that the reasons for people to join a community are mostly because of users being fascinated by topics, a new algorithm is put forward. This algorithm takes both the relationship of the topics and the structure of networks into account. Experimental results show that the proposed algorithm can simultaneously take advantage of structural information of the network and the topic information of the texts of the social networks, and effectively discovery communities.In summary, this dissertation aims at the three key issues of social networks including the analysis and mining of individuals, special nodes and communities, and some technologies such as the calculation of the relationship between users and topics, critical nodes finding and communities discovery, which will have brilliant perspective on social network analysis and data mining, and have great theoretical significances and practical application values.