Sentiment Evolution Of Weibo Users And Correlation Analysis With Cyber Event

Nowadays,Internet platforms such as Weibo have marked a new era for the public to express their attitudes,opinions and sentiments on the government's policies or various social phenomena.To analyze those ideas or sentiments,which is helpful to observe and warn cyber events,has practical significance to promote the governance in cyberspace.Guided by the application demand of public opinion control,this thesis mainly discussed user's sentiment and public opinions evolution on Weibo.It focus more on the sentiment classification method,the sentiment evolution analysis method and the correlation analysis between user's sentiment and cyber events.This thesis includes some aspects as follows.Firstly,in view of the fact that the existing sentiment classification algorithms put particular emphasis on semantic word embedding while ignored the sentiment classification combined sentimental information and user's information,the accuracy of sentiment classification needs to be improved.The two channels text convolutional neural network(TC-Text CNN),which combines multiple features,is proposed.By improving the algorithm of the word embedding,the part of sentiment information is enhanced by How Net.And the new implicit sentimental features are constructed from on the basis of the user's attributes and behaviors,as a supplement to the traditional feature.Based on the above work the TC-Text CNN is proposed,so that it can learn semantic information,sentimental information and user's information at the same time.Experiments proved that our method is applied in sentiment classification tasks and achieves better effect.Secondly,considering that the existing sentiment evolution methods ignore the relation among sentiment,users and focal points,two analysis methods were proposed.One is the sentiment evolution analysis method based on community division.Another is a sentiment visualization method based on the constructing sentimental river word cloud atlas.This method divides the user community by clustering algorithm,and the differences and commonalities of sentiment among different communities in different periods are analyzed in stages.In addition,a method of constructing sentimental river word cloud atlas is proposed,which visually shows the process of sentiment evolving with the transformation of focal points.The experiment found that opinion leaders are more likely to hold positive sentiments than ordinary users,and the shift of event focus will affect the development of users' sentiments.Thirdly,in view of the insufficient study of the relevance between sentiment and cyber events in the existing research,a method of analyzing the correlation between sentiment and cyber events is proposed and implemented in order to find the relationship between them.This method quantifies the sentiment tendency,sentiment intensity,influence and harmfulness of the event,calculates the correlation coefficient among the indicators,and analyze the relationship between the sentiment and cyber events.The experimental results show that the more harmful events are,the more likely they are to produce negative sentiment polarity.This conclusion can provide an important reference for the early warning of group cyber events.