Chinese Text Sentiment Analysis Based On Deep Learning

With the explosive growth of Internet text information,it has brought more and more unstructured text data containing the author's emotions and opinions.Extracting the text of emotional information from these data,will greatly affect the network public opinion analysis,social opinion guidance,commodity support decision-making development.The text emotion classification refers to the classification of the emotions contained in the text to be analyzed.Traditional methods are based on statistical or rule-based approaches.With the increase of the amount of data,the demand for human and material resources is growing,but the classification effect is getting lower and lower,it cannot meet the requirements of large data age,we need to put forward new methods urgently.The sentiment classification of text occupies the pivotal position in the research of affective analysis.In the 21 st century of information explosion,the research on sentiment classification of massive data attracts many researchers.How to deeply study semantic information of texts,express semantic features accurately,Accuracy is the goal of the study.In view of the shortcomings of traditional machine learning methods that can not learn textual semantic information,this paper proposes a solution to the problem of textual affective classification based on shallow learning features,which enhances the expression of semantic information in texts and enhances the semantics of semantic information Learning and understanding skills.The model of this paper has been improved in the following aspects:1)The new activation function is used,which improves the convergence speed and generalization ability of the model and alleviates the problem of gradient disappearance.2)A new optimization function is used,which makes the iterative learning rate have a definite range after every offset correction in the training process.3)Adding Dropout layer and L2 regular thinning method to the model to improve the fitting ability of the model and make the model more refined.4)Use Max Pooling technology to extract the maximum value of local features.It can be seen from the experimental results of the IMDB English public comment data set and the product-based Chinese comment data set that the four-point improvement proposed in this paper is effective.In addition,this article do a comparative study of multiple groups of more important parameters in the model,the impact of these parameters on the model.