| In recent years,with the rise of the national economy level,the proportion of automobiles in residents’ travel is increasing.At the same time,due to the rapid development of the Internet,various automobile comment websites have appeared on the Internet platform,such as Eka Automobile,Automobile Home and Echelon.Customers usually use relevant websites to learn about the brand,style,performance and price of the car before buying it.After purchasing a car,customers often take the initiative to share the experience of purchasing a car on these platforms.As a result,these platforms have a large number of car review data.Major automobile manufacturers are paying more and more attention to the information mining and analysis of this channel in order to achieve the purpose of analyzing user needs,improving existing products and mining competitors’ information.After a lot of observation,it is found that the subjectivity and randomness of the automobile comments on the website generally exist,which leads to the inaccuracy of the emotional polarity of the comments and the topic sentences of the automobile indicators concerned by the manufacturer.Therefore,the research on automobile review data mining mainly includes two aspects: on the one hand,quickly identify the emotional polarity expressed by each comment,which is helpful for manufacturers to get accurate emotional tendencies of customers.On the other hand,extracting thematic sentences of specific automobile evaluation indicators concerned by manufacturers from short-text automobile brief reviews can help to improve the efficiency of key information retrieval for review data,improve product quality and realize personalized marketing for customers.Therefore,this paper uses the crawler technology to collect and mine the automobile comments data of more than 20,000 customers on the top five comprehensive ranking Ica automobile websites in China.Firstly,it establishes "emotional classification analysis" and "topic extraction analysis" for automobile review data as two kinds of exploratory analysis objectives.Secondly,statistical theory is used to process and analyze the acquired data and establish feature engineering,which lays a good foundation for improving the effect of data mining research.Then,relevant models are constructed for the two analysis objectives.Finally,the evaluation of the model and the output display of the research results are carried out.Specific research contents and relevant conclusions are as follows:(1)In emotional classification analysis,text data marking,data cleaning and feature selection are the first tasks.Secondly,three algorithms,XGBoost,Naive Bayesian and Regularized Logic Regression,are used to model the "positive" or "negative" emotion binary classification of user reviews.From the results of emotional classification analysis,we can see that the model based on regularized logistic regression algorithm works best when it is used as a classifier.(2)In the analysis of topic extraction,firstly,after sentence-cutting and preprocessing of the crawled text data,a multi-classification regularized logistic regression model is established.Secondly,for each test text,the category probability of each sentence is predicted.On this basis,combined with reasonable sentence stitching rules,nine related topic sentences are extracted,including "appearance","interior","space","comfort","fuel consumption","power","manipulation","four-drive" and "cost-effective".From the results of extraction,it is found that the accuracy of this topic extraction method is high.In theory,the analysis method based on automobile review data proposed in this paper has certain reference value in data mining research in other fields.In terms of application,this method has a great guiding role in consumer behavior analysis and precise marketing for major automobile manufacturers. |