Research On Microblog Sentiment Analysis System Based On Machine Learning

As the technology grows,it is an inevitable trend to change the conventional way of information release which is independent and scattered.With its low threshold for information release,easy and fast interaction,and popularization of the audience,Weibo has quickly replaced traditional information release methods within a few years.It becomes an indispensable tool for people to obtain information and express their opinions on a daily basis.In 2018,the average daily text volume of Weibo was 130 million.How to mine the user 's emotional tendency from massive data,so as to timely grasp the emotional trend of the people,maintain stability and maintain the brand image in a timely manner,has become a topic of great concern to governments and enterprises.Compared with the refined content of traditional news reports,Weibo text is controlled within140 words,while comments are less written,and are mostly reduplicated to express emotions,without context.Therefore,the thesis adopts a statistical model-based word segmentation method and a word vector model to convert text information,uses a linear discriminant analysis model to model the text,and outputs the sentiment tendency of the sentence,which is convenient for companies and governments to control the sentiment tendency of Weibo speech.The dissertation first collects Weibo and comment data,and then uses data cleaning,Jieba word segmentation,stop words,Word2 vec neural network model to generate word vectors and other methods to process the data.Then,by comparing the effect of the LDA classification model and the Stacking integrated learning classification model on the sentiment of the sentiment,the LDA model with better performance is finally selected to solve the problem of sparse and missing semantic data on Weibo and comments.Finally,the ASP.NET MVC framework is used to carry the grafana visualization platform for visual display.The experimental results show that the LDA model used in this paper has an accuracy of 89%,and the experimental results are better than the Stacking learning model with an accuracy of74%;the emotional tendencies of the comments predicted by LDA model is basically correct.In this paper,through the experimental analysis of Weibo data sentiment classification,build a public opinion sentiment analysis system,in addition to visual display of public opinion information,but also to provide users a chance to test the emotional inclination.The experimental results show that the system can monitor the public opinion of microblog in real time,and use linear discriminant analysis model to analyze the emotional tendency of public opinion effectively,which is helpful for the government and enterprises to grasp the emotional trend of public opinion quickly and improve the work efficiency.