Research On Video Barrage Text Based On Sentiment Analysis

The current era is the era of information explosion,where people generate a large amount of data every day and are also surrounded by various big data,passively receiving a large amount of information.Especially with the popularization of the fourth generation communication technology(4G)and the promotion of the fifth generation communication technology(5G),a large amount of data has been brought,and at the same time,videos can be transmitted more smoothly on the network and watched more conveniently by people.Due to the increasing demand for real-time comments on videos,barrage technology is gradually becoming popular in the video field.As more and more people use bullet screens,research on emotional analysis of video bullet screens becomes particularly important.Research can not only help users better understand videos,but also help the internet better supervise and manage public opinion,making the internet environment more beautiful.In this thesis,we first use web crawler technology to crawl the bullet screen data of all TV dramas on Bilibili website;Then,the overall situation of the barrage data was explained in detail through charts and graphs;Finally,part of the barrage data was selected through stratified sampling for sentiment annotation,and the data was divided into training,validation,and testing sets for subsequent model training.This article mainly uses the BERT model to model bullet screen emotions,in order to obtain a model for sentiment analysis of bullet screen texts.Unlike most sentiment analyses that ultimately result in two classification results,this article ultimately obtains three classification results,namely no emotional tendency,positive emotional tendency,and negative emotional tendency.On the basis of obtaining the accuracy index of the model,an analysis of the methods for obtaining the Precision,Recall,and F1 values of multi classification models has been added.At the same time,the final model results were compared under different parameter combinations.Then analyze the emotional analysis results of the test set data,explain different types of errors,and propose solutions.Then the naive Bayesian model and the random forest model are used to model the emotion analysis of the bullet screen data,and the results of the two models are compared with the results of the BERT model.Through comparison,it is found that the BERT model has a higher accuracy than the naive Bayesian model and the random forest model.Finally,use the sentiment analysis model trained by BERT to conduct sentiment analysis on TV dramas.The final emotional analysis model results perfectly fit the plot.At the climax of the plot,the number of positive emotional tendencies is much higher than the number of negative emotional tendencies;When the antagonists are arrogant,the number of negative emotional tendencies in the barrage increases sharply,while the data on positive emotional tendencies in the barrage decreases sharply.