Bullet Subtitle User Community Division And Behavior Analysis Based On Emotion

Nowadays,bullet subtitle video websites are gradually becoming popular among young people,and more and more bullet subtitle users are keen to express their attitudes and opinions through bullet subtitles under hot videos.Dividing communities by the sentiment in the bullet subtitle texts sent by netizens,and then analyzing the behavior within each community,which can achieve precise control of events and public opinion,and also provide certain guidance for the governance of cyberspace.This thesis mainly takes BILIBILI video bullet subtitle website as the research object.After analyzing the shortcomings of existing bullet subtitle users,the sentiment analysis of the bullet subtitle text and the community division method of the bullet subtitle users are carried out.The main work is as follows.Firstly,aiming at the problem that the sentiment analysis method in current research cannot accurately measure the sentiment of the bullet subtitle,which leads to the low accuracy of the sentiment classification,a sentiment analysis method of the bullet subtitle based on affective computing and ensemble learning is proposed.First,expand on the basis of the traditional sentiment lexicon,construct the Chinese bullet subtitle sentiment lexicon,and combine the information of other dimensions of the bullet subtitle,and propose a seven-dimensional affective computing method for the bullet subtitle.At the same time,based on the idea of heterogeneous ensemble learning,a method of sentiment analysis of the bullet subtitle is proposed.The experimental results show that the method can effectively judge the sentiment of the bullet subtitle,and the effect in practical application is excellent.Secondly,in view of the inaccurate or even impossible measurement of sentiment similarity between some users,and the poor stability of the K-means algorithm,resulting in poor quality of the divided user communities,an improved K-means community division method based on feature regularization is proposed.First,regularize user sentiment.On the one hand,if the user's sentiment calculation is inaccurate,the overall sentiment at the current time is used to correct it;on the other hand,if the user's sentiment is missing,the missing sentiment is supplemented by the sentiment compensation method,and high-quality user sentiment representation is obtained.Finally,the problem of instability of K-means algorithm is solved by improving the selection method of initial cluster centers.The experimental results and visualization show that through feature regularization and improved community division algorithm,higher-quality communities can be effectively divided.Thirdly,by combining the research on sentiment analysis and community division,we designed and implemented a bullet subtitle user community sentiment analysis prototype system.The system can use user sentiment to analyze community behaviors and provide an effective means for realizing public opinion governance within the community.