Research On Feature Analysis Of Hype Accounts And Detection Of Hype Groups In Microblogs

With the rise and development of Microblog, the hype accounts suddenly rise as a new force in recent years. In order to seek illegitimate interests, they use illegal means to carry out the network public relations activities and deliberately incite users’ emotions, which has seriously disturbed the normal order of the Internet, making them becoming a significant force on the Microblog platform. As the hype models which hype accounts participate in are different, and there’re strong concealment and organization between hype accounts, which makes it difficult to identify these accounts for the traditional detection mechanism. In view of the above problems, the hype accounts in Microblog are carried out in-depth research in respect of hype accounts detection, hype groups detection, and role analysis of hype accounts. The main work of this paper is as follows:1. To solve the key problems in the research of hype accounts, a research framework for hype accounts is proposed. The related concept of hype accounts and the basic process of hype accounts research are given in this framework, which lays a foundation for further research.2. To solve the problems that hype accounts are highly concealed and impalpable, a method based on feature analysis for the detection of hype accounts is proposed. The features of hype accounts are analyzed from many angles, and a features database is constructed in this method, then the hype accounts are automatic classification by using several classification algorithms in data mining. Experiments show that this method is suitable for the detection of most hype accounts, with the accuracy rate of 95%.3. To solve the problems that hype accounts are strongly organizational and planning, a method based on mining maximum frequent itemsets for the detection of hype groups is proposed, which can quickly and effectively mining the groups who often participate in the spread of hype microblogs. The problem of hype groups detection is transformed into the problem of mining maximum frequent itemsets in this method, and a new algorithm(IIA) based on iterative intersection is proposed to improve the efficiency of mining maximum frequent itemsets. Experiments show that this method can find highly concealed hype groups with high accuracy.4. To solve the problems that the organizational relationships in hype accounts are quite hidden, a method based on IK-means for analyzing the role of hype accounts is proposed. Both of the individual attributes and network structural attributes of hype accounts are combined to extract feature vectors in this method, and a new clustering algorithm named IK-means is proposed to cluster the hype accounts accurately and divide them into different roles according to the clustering results. Experiments show that this method can discover different roles of hype accounts, which provides reliable basis for clearing the organizational structure of hype accounts.