Research On Sentiment Analysis Of Reading Software Reviews Based On BERT And LDA

With the progress of society and the development of information technology,the number of digital readers is gradually increasing.Compared with paper books,digital reading is more convenient,not only convenient to carry,but also readers can use their spare time for fragmented learning.Therefore,more and more people become loyal users of reading apps,a large number of reading apps emerge as the times require,and software reviews reflect users' attitudes and opinions on software features.Therefore,mining these fragmented and unstructured data can help users understand the software and developers better understand users' needs.This paper analyzes the user reviews of moujiang Novel Reading,moumao Nov-el,mouqie Novel and mouxin Reading,constructs the positive and negative dichoto-my model of emotions,and then mines the potential topics of reviews combined with LDA theme model,so as to compare the advantages and disadvantages of the four softwares and provide feasible suggestions for the development platform of reading software.The main analysis process is as follows:Firstly,the four software reviews are respectively deduplicated and cleaned,and the stars are scored according to the corresponding reviews,using a stratified sampling method Select some reviews from the four software,match the expanded emotional dictionary to calculate the emotional score,mark the comments with the emotional score based on the emo-tional score,and then combine manual screening to leave comments with obvious emotional tendencies.Secondly,the emotion classification model is constructed,the comments marked with sentiment tendency are divided into training set and test set.The training set is used to fine-tune the downstream tasks of BERT,and the senti-ment classification model of BERT is constructed.Then the model is compared with support vector machine classification model and random forest classification model.The accuracy of test set is used to compare the performance of the model,and the sentiment tendency of four software reviews is predicted by the model.Thirdly,the advantages and disadvantages of the four software features are compared.Based on the emotional classification results of Bert model,the LDA topic model is used to analyze the positive and negative comments,and the user attitude evaluation of the four software features is calculated.After the research and analysis of this paper,the BERT model of the emotional two-classification of reading software was finally constructed,with the classification accuracy of 82.73%.According to the classification of the emotional tendency of the four software reviews,namely moujiang Novel Reading,moumao Novel,mouqie Novel and mouxin Reading,the result is that mouqie Novel is more popular with users,with the most positive evaluation,and moujiang Novel Reading with the most negative reviews.According to the results of LDA theme model,the software is recommended to users and give suggestions to software developers in terms of software interface,novel content,listening function and so on.