Research On Automobile Customer Satisfaction And Product Decision Method Based On Sentiment Analysis

With the rapid popularization of the Internet,various automobile portal websites have become an open platform for many-to-many information exchange among automobile users.It is an important channel for consumers to understand the real driving experience and word-of-mouth evaluation of automobiles.It also subverts the traditional way that auto companies use questionnaires and visit surveys to track changes in market demand and obtain customers' feedback.However,how to mine the application value contained in the user-generated comment data and efficiently process these massive text data,provide a reference for consumers to purchase automobile products,and provide brand-new means for automobile companies to grasp consumers' needs.Based on this background and purpose,this paper conducts research and uses sentiment classification method to mine automobile review texts.The main research work is as follows:(1)Firstly,the crawler tool software was used to obtain the automobile review text data on each automobile portal website,and the data was cleaned for the problem of data noise and format inconsistency,so that the data format was standardized.The professional terminology dictionary for each performance dimension of the automobile was expanded and perfected,and a multi-dimensional de-redundancy algorithm was designed based on the dictionary to automatically classify every comment into the corresponding performance dimension.(2)Secondly,established the binary classifier to classify the sentiment of the review text,and maps the probability distribution of the two-class sentiment to the measurement of automobile customer satisfaction.The credibility of this method is highly dependent on the classification performance of the classification model,while the performance of existing the binary classifier based on machine learning or deep learning varies.In view of this,this paper proposed a method of model combining named parallel double station.Specifically,based on the existing sentiment classification methods,two classic and feasible binary classifier models(Convolutional Neural Network and Naive Bayes)were selected as a benchmark sentiment classification model for two parallel stations respectively,then the sentiment tendency of each comment output by the classifier on the two parallel stations was checked,and a consistent comparison was made.The text with the same sentiment tendency were retained as effective text to participate final marking of customer satisfaction.The verification results on the test set showed that,compared with a single classifier,Parallel dual station sentiment classification system proposed in this paper can further improve the precision,recall and accuracy of sentiment classification,thereby improving credibility of the sentiment score(customer satisfaction),while not causing a significant reduction in data volume.(3)Finally,based on the automobile customer satisfaction quantified by sentiment analysis,a consumer car purchase decision model was constructed.AHP was used to determine consumers' individual automobile performance preference weights,and TOPSIS was used to determine the candidate automobile ranking based on customer satisfaction.