Sentiment Analysis Of Car Review Text Based On Multi-label Classification

In recent years,with the booming development of automobile and Internet industries,more and more auto review data are left by users,and practitioners in related industries are paying more and more attention to the utilization of auto review data.At the same time,the continuous development of deep learning technology also makes the research of car review data more and more practical.In this paper,the attribution-level sentiment classification is carried out based on the text of automobile comments in real scenes.The main research contents and contributions of this paper are as follows:(1)Build a data set of automobile reviews that is in line with the actual scenario.A large number of automobile review data were collected from "Car Home",a website that provides automobile information,and the data were sorted out and re-annotated according to the research content of this subject.(2)Identify the comment attribute in the car comment text.Recognition of comment attributes in car comment text is essentially a problem of multi-label classification.Based on this problem,this paper proposes three solutions,respectively: The multi-label classification problem was transformed into several binary classification problems,and a loss function was defined based on the Circle formula to help the multi-label classification task to carry out back propagation during model training.The multi-task learning method was used to solve the multi-label classification problem.Then,based on the Text CNN model,this paper used the above three methods for modeling,and found that the multi-tasking learning method was significantly better than the first two methods.(3)Sentiment analysis of specific attributes is carried out according to the recognition results of comment attributes.In this paper,Bi-LSTM model and attention mechanism are used to conduct sentiment analysis on specific attributes of car review text.The experimental results show that compared with other commonly used sentiment classification models,the car attribute sentiment analysis model used in this paper has a better performance in the classification index,and can effectively classify the positive and negative emotions of each attribute in the car review text.