Emotional Analysis Based On Text Comments

The development of the Internet has not only changed people’s lifestyle,but also accelerated the combination of variety shows and the Internet.Compared to TV variety shows,online variety shows have diverse forms and richer themes.However,viewers feel more or less inadequate in the process of watching the program,and only continuous improvement can meet the diverse requirements of the audience.Therefore,from the perspective of viewers,this article uses the LDA theme model and machine learning methods to deeply explore the main concerns in audience reviews,and analyzes the emotional polarity of each aspect,providing suggestions for program producers to guide the development of online homemade variety shows in a better direction.Based on the audience reviews of 156 online homemade variety shows played on Douban.com from 2017 to 2021,this article explores emotional tendencies under different themes.The main research content includes:(1)data collection and preprocessing:using the "Octopus" crawler software to crawl audience reviews of 156 online homemade variety shows on Douban.com,and preprocessing them with de duplication,word segmentation,and de stop words.(2)Training naive Bayesian model,support vector machine model,decision tree model,and random forest model to classify emotions,and comparing and evaluating the performance of the four models,it is found that random forest has the best classification effect for this dataset,the overall accuracy rate reached 88%,and the positive and negative review precision rates reached 91%and 89%,respectively,which are the highest among the four models.(3)The LDA topic model is used to extract the four topics in the review and the attribute words contained in the corresponding topics.Then,the review text is divided into short sentences according to punctuation marks,and matched with the attribute words to obtain a set of short sentences under different topics.Finally,a trained random forest model is used to conduct emotional analysis on the short sentence set,obtaining the frequency of positive,neutral,and negative emotional tendencies,and displaying them visually.(4)In order to deeply explore the emotional tendencies of different types of online home-made variety show in different aspects,the random forest model was used to classify the emotions with the review data of competitive shows and reality shows as samples.It was found that the classification accuracy of competitive shows reached 0.90,negative emotions accounted for the majority,and the accuracy of reality shows reached 0.89.The overall emotional tendencies were positive emotions.(5)Finally,some suggestions are provided for the future development of online self made variety shows.For example,program producers establish cooperation with the review platform to directly understand audience feedback,and regularly conduct text mining and improvement on audience reviews;The comment platform sets more detailed comment options,such as "program link","program guest",etc.