Fine-grained Emotional Analysis Of News Commentary Based On Capsule Network And With The Help Of Multi-source Data

With the expansion of the popularity of the mobile Internet and the intelligent terminal equipment,newsreaders have more and more access to get news information,in this context,mobile news client has become one of the main channels for readers to get news.Compared to newspapers,television and other traditional news media,its advantage is that the news media and readers interact more closely.For example,after reading the news,readers usually give them feedback by making comments,meanwhile,the media will also conduct voting and other activities to obtain readers' emotional and attitude.In order to help the news media to get readers' emotional feedback more easily and to dig out the value contained in readers' comments as much as possible,this paper hopes to find an effective way of automatic emotion mining.There are a variety of methods for text emotion analysis,from relying on emotion dictionary and syntax structure,to machine learning method relying on manual feature selection,to deep learning method realizing automatic feature extraction,text emotion mining is becoming more and more efficient and accurate.This paper analyzes and summarizes the current research status of readers' emotion analysis.Combined with the newly proposed Capsule Network model in the field of deep learning,this paper makes an in-depth study on the method of automatic identification of readers' emotion caused by news.The main work includes:First,In this paper,the traditional binary classification method of "either praise or depraise" is abandoned,and the commonly used classification method of text emotion and the emotion distribution characteristics of real news comment data set are consideredcomprehensively.Finally,reader comments are divided into four categories: "anger","sadness","support" and "other".Second,this paper integrates the three data sources of reader comments,news reports and the number of likes of "hot comments" corresponding to all kinds of emotions,with the help of multi-source data to achieve a more comprehensive and accurate mining of reader emotions.Thirdly,the capsule neural network originally used in the field of image recognition is applied to the natural language processing task,the network structure is adjusted,the dynamic routing process in the model is improved according to the similarity measurement method of text data,and the classification effect is optimized.This paper conducted a comparative experiment on the real news comment data set,and compared the differences in the effect of the analysis method of this paper and the traditional classification model,the results show that the fine-grained emotion analysis model of multi-source news commentary based on Capsule Network can identify the reader's emotion more accurately than other existing methods.