Study And Implementation For Micro-Blog’s Network Community Construction And Opinion Leader Mining

Nowadays, as a new public opinion carrier and route of transmission, micro-blog is playing a more and more important part in public opinion emergence and transmission, and no other public opinion carriers can replace it. There’re many new characteristics in micro-blog comparing with traditional blogs, forums and goods comments. Three of them are highlighted:little obstacle, short and instantaneity. So it’s necessary for network community construction and opinion leader mining in micro-blog.In this thesis, research work can be divided into two parts:network community construction and opinion leader mining based on users’ tags, and network community construction and opinion leader mining in hot topics. For opinion leader mining, the thesis proposes two micro-blog opinion leader measure models, that is micro-blog user opinion leader measure model and micro-blog hot topic micro-blog opinion leader measure model.Firstly, the thesis constructs network community in micro-blog users by user tags, and users in the same community has similar tags which illustrate they have similar interests or identities and so on. This model measures micro-blog user influence by audience attention degree, user active degree and user value. Finally Analytic Hierarchy Process is made use of to assign weights.Secondly, the thesis constructs network community by micro-blog’s different emphasis on one topic, and micro-blogs in the same community have the same emphasis on this topic. Through classifying micro-blogs, network community is constructed, a classification method based on class feature words making use of LDA semantics information is proposed, and its result is compared with the classification methods based on SVM. The model of measuring micro-blog opinion leader for micro-blog hot topic measures micro-blog influence by explicit influence, implicit influence and user influence. Finally Analytic Hierarchy Process is made use of to assign weights.The experiment results show that two micro-blog opinion leader measure models proposed in this thesis can discover opinion leaders, and the classification method based on class feature words is better than the method based on SVM in micro-blog classifying requirements in this thesis, but for spam micro-blog reduction, the method based on SVM can work better.