Research On Sentiment Analysis Of Chinese Text Based On Deep Learning

The continuous progress of Internet technology has driven the rapid development of self-media.Self-media provides users with a platform for expressing opinions and expressing personal emotions,such as Weibo,Facebook,Twitter,etc.Internet has accumulated a large amount of text information that have some personal opinions and emotional tendencies.By analyzing the emotions and opinions contained in these texts,we can get the crowd's emotional changes and the relationship between emotions for specific events,it plays an important role in obtaining public opinion guidance and product's social evaluation.Text sentiment analysis,also known as text opinion mining,is a basic and important task in the field of natural language processing,especially for the extraction and classification of potential emotions,attitudes,opinions and other similar information in the text.In this paper,Firstly,the Chinese text sentiment classification of binary text was studied.a binary Chinese text sentiment analysis method based on BiLSTM-CNN serial hybrid model was proposed.Secondly,because the granularity of binary text sentiment analysis is too large to meet the needs of more realistic environments,the Chinese sentiment classification of ternary text is studied.According to the characteristics of natural language,we accurately understand the semantics of the text,such as the context,sentence structure of the text,local semantic information and part of speech of the text,etc.A sentiment analysis method for ternary Chinese text based on feature fusion is proposed.The main research work of this article is as follows:(1)Aiming at the problems of poor model generalization ability and inaccurate semantic understanding of text in the current text sentiment analysis methods and the accuracy rate of text sentiment analysis models is not high,etc.A method of binary Chinese text sentiment based on BiLSTM-CNN serial hybrid model was proposed.The method first uses the Word2 Vec vocabulary vectorization tool to convert the text vocabulary into real vectors of the review text,and secondly the context information is extracted by using Bi-directional Long Short Term Memory(BiLSTM)from the text;and then,the local semantic features is extracted according to the Extracted context information by using Convolutional Neural Network(CNN).and finally uses Softmax to classify the personal sentiment tendency expressed by the text.Through experimental verification,this method can effectively improve the performance of text sentiment analysis to a certain extent.(2)Aiming at the problems of current text sentiment analysis methods,that a wordscontain multiple semantics,and the semantic information expressed by the text cannot be accurately understood,resulting in poor text sentiment analysis performance and low accuracy,etc.A ternary Chinese text sentiment classification method based on feature fusion is proposed.This method firstly uses the GloVe vocabulary vectorization tool to map the discrete text vocabulary that microblog data set with the new coronavirus theme to real number space,and combine parts of speech into this process when converting the text vocabulary into a real number vector containing certain semantic,thus solving the problem that a word contains multiple semantics;Secondly,using TextCNN to extract local semantic information of the text from different fields of view,using introducing self-Attention mechanism in BiGRU to extract global semantic feature from the perspective of sentence structure and context information;and then feature fusion of local semantic information and global semantic information;Finally,Softmax is used to classify the text sentiment,and good text classification effect is obtained on standard Weibo dataset.