Research On The Prediction Of Potential Opinion Leaders In Social Media

Today's society is experiencing rapid development of science and technology.As general secretary xi jinping said in his speech,"promoting integrated development of media and building all-media have become an urgent course for us.Among the numerous network social platforms,microblog is an important virtual social platform for people to socialize,express opinions,obtain information and talk about topics.Microblog has a huge user resource,and the network public opinions generated in microblog also have a huge impact on our real life.Therefore,under the situation of tense online public opinion,it is particularly important to monitor online public opinion,and opinion leaders and potential opinion leaders play a crucial role in the evolution of public opinion.This paper will conduct research on the prediction of potential opinion leaders in weibo.In the experimental method,we use the machine learning algorithm to complete the experiment.Five characteristics,namely,personality characteristics,theme characteristics,text characteristics,microblog quality and microblog topic concentration,were selected as user tags.The big five personality model is selected for the personality characteristics.The model is widely used and has a complete system,including extroversion,openness,agreeableness,responsibility and neuroticism.According to the classification of topics in the background of microblog,18 kinds of topics such as tourism,photography,film,beauty makeup,music and education are selected as the theme features of users.Since microblog texts are short texts,LDA theme model is selected in this experiment to extract the theme of users' microblog.At present,the research on opinion leaders or potential opinion leaders starts from the phenomena such as user social relations or surface data.Based on the definition of opinion leaders in communication science,this paper studies the essential characteristics of potential opinion leaders and does the following:It redefines potential opinion leaders,which specifically refers to users with potential and characteristics to become opinion leaders even though the number of followers,comments or retweets has not reached the standard of opinion leaders.The role of potential opinion leaders is to promote or guide information or public opinion guidance after opinion leaders.The microblog data of opinion leaders are obtained through the search engine of microblog users.According to the distribution of the number of microblog comments in time,the microblog text data of users before they become opinion leaders are determined.Pre-selected 18 types of weibo theme tags,and extracted the themes from users using LDA theme model.A questionnaire survey was conducted on some weibo users by using the big five personality scale.Combined with manual labeling,the user's personality was identified according to the basic attribute information and the characteristics of weibo posts.Through six classification algorithms: decision tree,KNN,GBDT,Xgboost and the random forest integrating learning classes,the classification model is trained to predict the user's personality traits.Finally,the KNN classification model has the best classification effect and the accuracy result is 0.88.Through the statistics of users' microblog topics,the number of weibo posts of various categories in users' microblog topics is obtained,and the value of users' concentration on the topic selection is calculated.The smaller the value is,the more concentrated users are in the topic selection,and the more prominent the topic category is.Will collect user weibo,according to the original plain text microblogging,original graphic weibo and forwarding microblogging three types of classification,statistics for all kinds of weibo,combined with the opinion leaders of the number of forwarding microblogging,additional weight to these three categories,the quality of the calculated weibo user value,the higher the quality showed that the bigger the attractive to other users.According to the characteristics of the five categories of weibo users,the three classification models of decision tree,KNN and SVM are adopted.The training set is divided into four categories according to the presence or lack of personality and topic tags.The results show that the best classification effect is achieved when the personality and subject labels exist simultaneously.The F1 value of SVM classification model is 81.13.In the absence of personality and subject labels,the worst classification effect is achieved.This proves that there is a correlation between personality and thematic characteristics,which opens up new ideas and lays a solid foundation for future research on potential opinion leaders or opinion leaders.