Subject Classification And Sentiment Analysis Based On Automobile Reviews

With the rapid development of science and technology,internet has become an indispensible part of human life,and consumers’ comments on the Internet have been growing exponentially.Online reviews on automobiles,which contain users’ comments on various aspects of automobile products,have been proven to be a valuable resource.Such comments may play an important role in guiding their purchase decisions.For automobile manufacturers,comments from buyers may help them find out problems and understand users’ needs in a timely manner,so as to improve automobile product design.However,there is lack of research on online automobile reviews,and the application of the popular deep learning models,are even more scarce in this field.Thus,this paper attempts to apply the deep learning model to conduct sentiment analysis with online automobile reviews,which may contribute to expanding the application scope of the deep learning model to a certain extent,and enriching the theoretical research on sentiment analysis of online reviews.This paper uses Python software to obtain data on online automobile reviews of seven major topics,namely power,space,interior,configuration,comfort,appearance and fuel consumption from Pacific Automotive Network.The text convolutional neural network model(TextCNN)and the bidirectional long and short-term memory model(BiLSTM)were constructed for experiments.Finally,the optimal model was obtained after parameter debugging.In addition,this paper obtained emotional comments on automobiles from three websites: Pacific Auto Network,Aka Auto,and Car Quality Network.By constructing BiLSTM-CNN hybrid neural network model,with the aim of analyzing the car emotional comments,the reliability of the model is verified through comparative experiments.The results show that(1)Based on TextCNN,topics of automobile reviews are classified with seven categories.By comparing the classification results of single convolutional kernel and hybrid convolutional kernel,it could be concluded that the smaller the convolutional kernel is,the better the classification effect is and the combined convolutional kernel is better than the single convolutional kernel.When the dropout parameter was tested,it could be concluded that the classification was the best when the Dropout value was 0.5.In addition,this paper also uses BiLSTM to classify the topics of automobile reviews,and the results proved to be better than the classification with TextCNN.(2)By constructing the BiLSTM-CNN hybrid neural network model to conduct sentiment analysis on positive and negative evaluations of automobiles,and comparing with a variety of single models,the paper verified the feasibility of the BiLSTM-CNN hybrid neural network model in the field of sentiment analysis.The dichotomy and trichotomy of text data have been popular in previous literature.In this paper,the obtained reviews are divided into seven categories,and TextCNN and BiLSTM are used for multi-classification research,which may fill the gap in this field to a certain extent.At the same time,a hybrid neural network model of BiLSTM-CNN with better classification effect is constructed,which is also of practical reference value to automobile manufacturers and potential car buyers.