Research On Classification Of Network Group Events Based On The Social Tagging

With the development of the network, social tagging technology has been widely used,making the mass media and interpersonal methods undergoing revolutionary change. Usersannotate the text through the micro-blog and social networking sites, causing an increasingnumber of network group event. Network group event early warning and response haveattracted the attention of society, government and academia. The topic type of the networkgroup event is the key of network group event early warning, which is a preliminaryjudgment, recognition and evaluation of network group event. The traditional classificationis the top-down, standard single method, which is lack of the collective wisdom. Thereforein the new network environment the topic classification of network group event is a hotspotresearch in recent years.In this thesis, the topic classification of network group event based on social taggingmainly includes two points:(1) First of all, analyzing the data set of social tagging is based on formal conceptanalysis. The data set contain the users, resource and tags. But the characteristics ofopenness and freedom, which are lack of hierarchy, cause the tags flat, influence theefficiency of the resource and can’t accurately identify the semantics resource. The paperanalyzes the data of social tagging from two aspects of the behavior of tagging and thepreference, establishing concept lattice. In the research of behavior of tagging, excavatingthe leader and the group of users by calculating the number of the linked node and the depthof the user-tag node. In the research of the users’ preference, finding the resource ofpreference by calculating the frequency of the linked node and the position of theuser-resource node.(2) In view of the traditional classification algorithm of dimension reduction, datasparse, lack of semantic, this paper puts forward the LDA model of text method based onthe tags. In the original three layer structure of the LDA model, introducing the tag layer todig out the topic of text. According to the model, the topic distribution, the word distributionand the tag distribution are inferenced by Gibbs sampling algorithm. First, analyzing thematrix of topic-word with the principle of “big to small”, filtering the words which is closeto or equal to zero, so as to reduce the dimension of words and calculate the similarity of thetopic. Second, analyzing the matrix of tag-topic and viewing the topic as the characteristic vector of the tag, calculating the similarity of the tag. Last, in view of the weight of tags,analyzing the matrix of topic-text and establishing the classifier of text, which verify thefeasibility, accuracy and superiority based on the tag of the LDA model.