Research On Classification Of Comment Text Sentiment Based On Deep Forest Learning

With the advent of the Internet era,the number of Internet users in China has been growing rapidly every year.In particular,the rapid development of social media such as weibo and Meituan has brought great changes to people's lives.Users can express their opinions freely through these new carriers,and text plays an important role in this process.Therefore,emotional analysis of texts has become a hot topic among researchers.At present,there have been a lot of researches on coarse-grained text affective analysis at the text level and sentence level.However,in the face of the similar text of commodity review,coarse-grained text emotion analysis cannot obtain the specific object of emotion.Therefore,from the perspective of fine-grained text emotion analysis,this thesis analyzes the text's emotional orientation based on the deep forest model.The main research contents of this thesis are as follows:(1)Extract text features.This thesis extracts two kinds of text features,which are binary feature and affective semantic probability feature.Firstly,based on lexicon and syntactic dependency analysis,this thesis extracts the evaluation object--evaluation word polarity of the text by combining six proposed rules.Then,the evaluation objects are clustered to complete the extraction of binary features.Finally,in order to make up for the shortcomings of binary features,this thesis extracts the emotional semantic probabilistic features of the text by combining dictionary and semantic dependency analysis,and normalizes the included features to improve the classification efficiency.(2)Improve the classification model.Firstly,this thesis improves the representation learning structure of the deep forest model cascade layer,enhances the representation learning ability of cascade forest,and solves the problem of feature information weakening and feature dimension enlargement in the process of training.In addition,considering that different text features may have different impacts on the classification results,this thesis proposes to combine the integrated learning algorithm with the deep forest model to obtain a new model BFDF(the BoostingFeature of Deep Forest),so that the model can notice the importance of features in the training process and adjust the weights of features accordingly,so as to obtain a better classification effect.Finally,the effectiveness of the proposed method is proved by experiments.(3)Implement the text classification system.A text emotion classification system based on deep forest learning is designed and implemented.The system's main function is to achieve the training of the model,and forecast the comment text emotional categories,still can adjust the parameters of the model according to the results of the prediction,so as to obtain better classification performance of the system.