The Research On Multi-Classification Of Emotions Based On Chinese Micro-blog Text

With the popularity of Internet applications,the public has commented on the Internet platform,and the frequency of participation in social event discussions has increased dramatically.As one of the popular social network platforms for domestic Internet information exchange and sharing,Weibo's remarks can reflect the social sentiment tendency of the domestic people to a certain extent.Therefore,it is very meaningful to mine and analyze Weibo text data and classify Weibo text emotions.The emotional classification of microblog text is to effectively classify the emotional categories of text in Weibo.Text data representation methods and classification algorithms have a greater impact on the accuracy of text classification.Since the word vector obtained by the traditional language model training does not contain emotional features,and for different word vectors,there are also differences in various text sentiment classification algorithms,which are not good for the application of microblog text sentiment class ification.In response to these problems,the research work carried out in this paper is as follows:1.The word vector obtained by training for the Word2 Vec model does not have the emotional feature.This paper proposes to integrate the emotional polarity of the network sentiment dictionary into the word vector to expand the gap between the emotional word vectors.And by combining different classification algorithms for emotion classification,to obtain higher emotional classification accuracy.The experimental results show that the classification accuracy is improved after the word vector is integrated into the network sentiment dictionary polarity.Among them,the accuracy rate combined with the support vector machine reaches 91.27%,which is 2.54% higher than the accuracy combined with logistic regression.2.In view of the phenomenon of polysemy in Weibo text,this paper uses BERT model to solve the problem,which eliminates the influence of polysemous words.At the same time,the model carries out the word vector training in words,and uses the emotional label of the text to emotionally label the words,thus solving the problem that the unaccompanied words have no emotional information.In this paper,the word vector obtained by the BERT model is used as the input of the deep learning classification algorithm to classify emotions.Experiments show that the accuracy of emotional classification is 92.66%,which is higher than the method of emotionally categorizing emotional polarity into word vector.