Research On Sentiment Classification And Evaluation Of Car Reviews Based On Evidential Reasoning

With the rapid development of the Internet and the social economy,people’s lifestyles and consumption habits have undergone profound changes.An increasing number of people choose to buy cars and share their own experiences of car buying and product using on relevant car review sites such as Pacific Auto.These reviews have become an important basis for consumers to understand car information,for companies to meet users’ demands.How to quickly and effectively use the users’ reviews on these car review sites to mine the emotional information contained in them and consumers’ demand for products is an important issue at present.Therefore,this paper collects massive car online reviews as data support to carry out evidence-based reasoning of car review sentiment classification and car comprehensive evaluation analysis,so as to assist consumers to make more accurate decisions,and help companies more easily and accurately extract users’ demands from car reviews.The specific work content is as follows:(1)This paper uses python crawler to obtain online review data of various car review attributes from the Pacific website,and performs preliminary cleaning work such as deduplication and removal of garbled characters.In the process of word segmentation,constructing an improved word segmentation dictionary by considering the professional vocabulary and online popular words in the automotive field,and the stopwords vocabulary of Harbin Institute of Technology,Baidu and Sogou are integrated to remove invalid stopwords,thereby improving the quality of word segmentation.Based on this,the chi-square test method is used for feature extraction,and the improved TFIDF method is used to calculate the weights to construct the text data vectorization matrix.(2)This paper proposes a multi classifier based on evidence reasoning car review sentiment tri-classification method,which uses naive bayes,logistic regression and support vector machine as the basic classifiers,integrates three base classifiers with evidence-based reasoning rules,and considers the weight and reliability of each base classifier,then the sentiment text is divided into three categories.Compared with the single base classifier and the popular integrated classifier,the method proposed in this paper has better effective classification(3)On the basis of the three-class sentiment model constructed in this paper,a comprehensive evaluation analysis of cars based on evidence reasoning is carried out.Taking SUV models as an example,the sentiment distribution of each review angle of the car and the comprehensive sentiment distribution are analyzed,and the advantages and disadvantages of automobile are mined by word cloud method.At the same time,the process is applied to initially build the system framework of automobile comment emotional mining system,which is convenient for enterprises to understand the analysis results directly.