Research On Chinese Sentiment Classification Model Based On Deep Learning

In today's highly convenient Internet,Chinese netizens' enthusiasm for participating in the Internet is unprecedentedly high.Relying on various online social platforms and e-commerce platforms,netizens' comments have touched on all aspects of our lives.In the information age today,these comments contain information with important value.Sentiment analysis of these Chinese comments will help the government monitor public opinion and help e-commerce platforms improve service quality.Based on the deep learning network,building an efficient Chinese comment sentiment classification model has important research significance.Based on deep learning technology in sentiment analysis tasks,there are convolutional neural networks(Convolutional Neural Network,CNN)and recurrent neural networks(Recurrent Neural Networks,RNN).Among them,RNN network is applied because of its unique structural characteristics.And this is because the unique loop structure of RNN can integrate the influence of the input at the current moment and the output at the previous moment to determine the output of the current node.It is precisely because of this that it can fully learn the difference between the text and the text.In the past period of time,the word embedding technology represented by Word2 Vec is a core representation technology in the related field of natural language processing research.It can transform words and words into words through unsupervised learning.A dense real number vector containing word meaning information.The BERT model and the multi-head attention mechanism used in the Ro BERTa model can easily solve the problem of ambiguity,and the Ro BERTa model used in this article has made many improvements on the basis of the BERT model,and the overall effect is better than the BERT model and the model is more robust.In this paper,through the analysis and application research of the Ro BERTa model,this latest deep learning network model is applied to the Chinese comment sentiment classification task,and in view of the limited amount of data,the method of data enhancement is used to improve the experimental effect.The main research contents are as follows:(1)Research on sentiment analysis model based on Ro BERTa.The structure and principle of the Chinese sentiment analysis model combining Word2 Vec with CNN,RNN and its variants,which was more popular in the past few years,was analyzed.The structure and principle of the BERT model and its improved Ro BERTa model were also analyzed.A comparative experiment was carried out on the two data sets,which confirmed that the Ro BERTa model constructed in this article works best.(2)The influence of the amount of data on the classification effect of the model is studied.In the original two experimental data sets,it is found that for the same model,the overall experimental evaluation index for the large amount of data is better than the overall experimental evaluation index for the small amount of data.So another data set is used for verification.First,this data set is divided into blocks,and20% of it is taken as the test set,and then the increasing amount of data is taken out of the remaining data as the training set,and compared on the Ro BERTa model.Finally,when the test set remains unchanged,the larger the training set data volume,the better the classification effect of the model.(3)The impact of data augmentation on model classification is studied.Use EDA(Easy Data Augmentation)and back translation methods to expand the data set.Each method expands the training set data set to twice the original size,and the test set remains unchanged before and after the expansion.Comparative analysis of experiments is carried out.The results show that the overall experimental results processed by the two methods are better than the original training set,and the better data enhancement method is back translation.