Topic Mining Of Weibo Comments Based On STM And Emotional Evolution Research

The value of data is more and more prominent in the digital information age,and more and more attention is paid to information contained in various fields.However,with the development of the times,data is no longer a simple numerical numeric data.More and more unstructured text data are gradually flowing into people’s view.From commodity reviews to job search information,or news bulletins,text data has changed the situation of traditional data analysis,and the analysis form of manual reading or expert scoring can no longer meet the needs of researchers.It has become an important research task to study unstructured data using the technology of text data analysis.Nowadays,the domestic and foreign computer fields have achieved satisfactory results in text data analysis,and applying it to other areas of research has become a requirement to conform to the development of the times.At present,there are widely used search heat index,public sentiment index and market sentiment index.The related data comes from the search volume of the portal website,social and economic news,and various kinds of community posts.However,little research has been done on such special public opinion data as Li Wenliang’s Weibo comments,a public figure who has passed away,Weibo is no longer updated,and community messages are constantly flooding the screen.Thus,this kind of comment data is taken as an example in this paper,and the mature topic mining algorithm in academic circles is used for reference to analyze Weibo’s comments.The descriptive statistical analysis method is used to make an overall overview of the data firstly after data preprocessing.Secondly,the traditional text classification method is used to classify the emotion of comments data,and the emotion polarity and emotion distribution of comments are analyzed by stages.Finally,three topic models: Latent Dirichlet Allocation(LDA)、Biterm Topic Models(BTM)、Structural Topic Models(STM)are used for topic mining of comments data to reveal different topic distributions.At the same time,the numerical results of emotion classification are used as binary covariates,and time variables and topic content covariates are added to build a linear model to analyze the evolution trend of topic popularity.The results show that:(1)The number of comments in a year,except for the large fluctuation on a specific date,gradually decreases with time and then tends to be flat but continues continuously,reflecting the particularity of the data.Except "Dr.Li",words such as "Good Night" and "Salute" appear frequently in the message content,and they are at the top of the high-frequency words on ordinary days and special days,respectively.Based on the results of high-frequency co-occurrence words,it is preliminarily concluded that the messages in the comment area are mostly blessing and daily communication.(2)When the comment emotion is classified into two categories,the number of positive comments is always higher than that of negative comments in each stage.In the multi-classification of emotions,emotions like "good" and "happy" are the main trends in each stage,followed by emotions like "sad".On the one hand,they show the enthusiasm of commenting on emotions,and on the other hand,they prepare data for the theme model.(3)Comparing the topic mining results of LDA,BTM and STM,it is found that there are limitations in both LDA and BTM models.Relatively,the results obtained by STM are excellent.On the one hand,it provides the excavated topic words,on the other hand,it reveals the distribution of different topics.The topics can be grouped into eight categories in conjunction with a manual reading summary,with comments such as "Communication: Sharing Life" and "Communication: Holiday Blessings" accounting for the majority of all comments.The emotional mining results of STM also show a more positive side.In addition,it also reveals the evolution of topic popularity in a year.Among the 12 topics significantly affected by time,the number of topics with initial popularity rising is lower than that with declining popularity.To sum up,from the perspective of statistical analysis methods and computer technology and psychology,this paper analyzes the development trend of public opinion behind public figures,which provides more perspectives for similar public opinion analysis in the future,helps to pay more comprehensive attention to social public opinion and provides a technical reference path for the government to control public opinion and make decisions.