Prediction Of Microblog Opinion Leaders Based On Topic Dynamic Characteristics

The activity,recognition and authority accumulated by opinion leaders through their long-term participation in information release and exchange have an important impact on the development of online public opinion.However,the existing research mainly determines the weight of indicators through questionnaire survey,and the selection of indicators such as forwarding number is based on the post-event data of a single event.The effectiveness of rapid prediction of opinion leaders in the early stage of the development of public opinion and guidance of public opinion through opinion leaders is questionable.Considering the dynamic characteristics of the growth of opinion leaders,this paper carries out prediction research on opinion leaders based on the dynamic characteristics of topics,extracts effective information from the massive data accumulated on the microblog platform,more comprehensively and accurately describes the professional authority of opinion leaders,and improves the accuracy of prediction of opinion leaders.The specific work of this paper includes :(1)Analyze the correlation between topics and the influence of opinion leaders.This paper selects three types of typical events to construct the coupling matrix of "Topic-Opinion Leaders".It is found that most opinion leaders have strong influence in their professional fields,and some of them have crossdomain development.Therefore,it is necessary to extract the dynamic characteristics of topics as prediction indicators of opinion leaders.(2)Constructing a prediction index system of opinion leaders based on the dynamic characteristics of topics.This article consider index measurability of data at the beginning of the event,based on the users’ personal attributes to build general opinion leaders prediction on the basis of index system,from the user personal radiation,user authority,user participation,user history influence dynamic characteristic,user topic to build a more complete in five dimensions,based on the topic of dynamic characteristics of the microblogging opinion leaders predictor.(3)Based on the dynamic characteristics of topics,a prediction model of opinion leaders is established.In this paper,a prediction model is constructed based on the general prediction index without topic feature and the prediction index of topic dynamic feature.The user index data is quantified as the input variable,and the support vector machine model,random forest model and artificial neural network model are selected to classify the prediction and evaluate the effectiveness of the prediction index system.The experimental results show that the prediction effect of the prediction indicators of microblog opinion leaders based on the dynamic characteristics of topics constructed in this paper is more than 90%,which is effective in classification prediction.The results show that the whole index based on the topic feature has higher prediction accuracy under the two models,and the support vector machine model has the best prediction effect.Historical re-rating data is an effective index for opinion leaders to predict,and the dynamic characteristics of the topic is an effective supplement to users’ static labels,which can dynamically depict the growth trajectory of users’ professional authority,and is one of the important symbols of their influence.