Research On Sentiment Analysis And Clustering Of Microblog Evaluations Of Crisis Events Based On BER

With the development of the Internet era,the power of information dissemination continues to strengthen,and in today’s booming self-media industry,every user can evaluate various crisis events on social networking platforms.This process,due to the information asymmetry,inconsistent views and different perspectives between users and users,users and platforms,users and official media is highly likely to lead to further polarisation of online information opinion on crisis events.As the largest social network platform in China,Weibo’s commentary data reflects the characteristics of users’ evaluation of information on crisis events.Based on this,this paper selects microblog comment data as the object of empirical research and analyses the emotional tendencies and clustering characteristics contained therein,so as to analyse the characteristics of microblog users’ evaluation of crisis events.This paper adopts python crawler technology to obtain 33 comments of popular blog posts under the "Zhengzhou 7.20 rainstorm" event,on the one hand,the data is analysed by sentiment dictionary,and the sentiment value of each comment is calculated by using sentiment formula.The BERT sentiment classification model was used to train and analyse the sentiment dictionary data and sentiment values,and then the model was used to predict the sentiment tendency of the microblog comment data;on the other hand,the sentiment comment data was subjected to text clustering,and its word frequency and clustering results were analysed.It can be found that the sentiment tendency of users was positive in the early stage of the crisis event,while the later stage of the event showed a high level of negative sentiment,while about one-third of users maintained a neutral sentiment tendency throughout the overall development of the event.Based on this,for the control of public opinion on crisis events,the platform needs to improve supervision and real-time monitoring;the government needs to respond in a timely manner and improve its credibility;and the users need to maintain a rational attitude and consciously screen information.