基于文本分类的新能源汽车满意度研究——以比亚迪与特斯拉为例

A Study on Satisfaction of New Energy Vehicles Based on Text Categorization—Taking BYD and Tesla as an Example

摘要: 量化评估新能源汽车满意度对于推动新能源汽车产业发展、未来能源转型具有重要意义。本文使用python爬取汽车之家上比亚迪与特斯拉两品牌的新能源汽车口碑评论数据，对非结构化数据进行LDA主题建模分析；对结构化数据提出一种基于Stacking集成学习的文本情感分类算法以了解各属性的满意度情况，在对比近十种同质算法后，选择精度最佳的Catboost、神经网络(ANN)和随机森林(RF)构建多基分类器，并利用逻辑回归(LR)作为元分类器，实现对汽车结构化维度的情感分类，最终准确度高达到95%。研究发现，购车体验、车型选择与噪音问题是结构化数据中被忽视的用户关注维度。而就满意度来看，比亚迪的整体满意度高于特斯拉，且均在内饰方面表现较差。未来应关注内饰部分的设计改进，以提升用户体验并改善用户满意度。

Abstract: Quantitatively assessing the satisfaction of new energy vehicles is of great significance for promoting the development of new energy vehicle industry and future energy transition. In this paper, we use python to crawl the review data of new energy vehicles of BYD and Tesla brands on Autohome.com, and perform LDA topic modeling analysis on the unstructured data; and propose a text sentiment classification algorithm based on Stacking integrated learning for the structured data to understand the satisfaction situation of each attribute. After comparing nearly ten homogeneous algorithms, Catboost, neural network (ANN) and random forest (RF) with the best accuracy are selected to construct a multi-base classifier, and logistic regression (LR) is used as a meta-classifier to realize sentiment classification of structured dimensions of cars, with a high final accuracy of 95%. It was found that car-buying experience, model selection and noise problem are neglected user concern dimensions in structured data. In terms of satisfaction, BYD’s overall satisfaction is higher than Tesla’s, and both perform poorly in the interior. In the future, attention should be paid to the design improvement of the interior part to enhance user experience and improve user satisfaction.