主管:中国科学院
主办:中国优选法统筹法与经济数学研究会
   中国科学院科技战略咨询研究院

中国管理科学 ›› 2023, Vol. 31 ›› Issue (11): 114-127.doi: 10.16381/j.cnki.issn1003-207x.2021.0446

• • 上一篇    

基于随机Kriging的汽车品牌质量序贯主题比较研究

施文1,渠玉杰1,蒋国银2()   

  1. 1.中南大学商学院,湖南 长沙 410083
    2.电子科技大学公共管理学院,四川 成都 611731
  • 收稿日期:2021-03-07 修回日期:2021-11-18 出版日期:2023-11-15 发布日期:2023-12-05
  • 通讯作者: 蒋国银 E-mail:jiangguoyin@uestc.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(71971219);国家社会科学基金资助重大项目(22&ZD128);湖南省杰出青年基金资助项目(2022JJ10084)

Comparative Study on Sequential Topics of Automobile Brand Quality Based on Stochastic Kriging

Wen SHI1,Yu-jie QU1,Guo-yin JIANG2()   

  1. 1.School of Business, Central South University, Changsha 410083, China
    2.School of Public Administration, University of Electronic Science and Technology of China, Chengdu 611731, China
  • Received:2021-03-07 Revised:2021-11-18 Online:2023-11-15 Published:2023-12-05
  • Contact: Guo-yin JIANG E-mail:jiangguoyin@uestc.edu.cn

摘要:

汽车产业是国民经济的主导和支柱产业之一,汽车质量安全问题备受关注。现有研究主要探索汽车质量的召回制度、召回产生的影响以及缺陷产品的发现,未发现从大规模质量缺陷的文本中挖掘汽车召回主题并指导实践的研究。本文从消费者投诉和汽车召回两个角度研究合资背景下汽车质量差异及不同汽车品牌在汽车质量各个维度的优劣势。本研究使用潜在狄利克雷分配(latent Dirichlet allocation, LDA)模型对汽车召回文本进行分析,由于消费者投诉文本呈现出较强的序贯相关性,本文基于序贯潜在狄利克雷分配(sequential latent Dirichlet allocation, SeqLDA)模型研究该类文本。考虑到SeqLDA具有较多影响拟合效果的输入参数,本研究提出随机Kriging(SK)和SeqLDA相结合的主题优化模型(SK-SeqLDA)对消费者投诉文本进行分析。研究结果表明:LDA适合于汽车召回文本,而SeqLDA适合于消费者投诉文本,SK-SeqLDA能较大幅度地增强序贯主题模型的拟合效果。另外,消费者投诉中多个主题涉及车身附件及内饰,汽车召回中多个主题涉及制动问题;进口汽车品牌召回主动性较强,主要集中于发动机;合资汽车品牌召回主要集中于车辆控制和制动踏板;自主汽车品牌各主题占比均衡。该研究方法和研究结论可为汽车制造商进行召回管理决策和进行消费权益保护等提供理论参考和方法支持。

关键词: 消费者投诉, 汽车召回, SK-SeqLDA模型, 非参Bootstrap检验

Abstract:

The automobile industry is the leading and pillar industry of the national economy, and the quality and safety of cars have aroused broad public concern. Most of the research has been conducted from the aspects of the automobile recall system, its impact, and the discovery of defective products. However, no investigations are found to mine the topics of automobile recalls from large-scale text data of quality defects. The differences and the advantages and disadvantages of different automobile brands in each dimension of automobile quality under the joint venture background are studied from the perspectives of consumer complaints and automobile recalls. Since the texts of automobile recalls do not show a strong sequential correlation, the latent Dirichlet allocation (LDA) model is used to analyze it. Due to the strong sequential correlation in the texts of consumer complaints, such texts are studied based on the sequential latent Dirichlet allocation (SeqLDA) model. Considering that the SeqLDA model has more input parameters that affect the effect of model fitting, the optimized model (SK-SeqLDA), which combines stochastic Kriging (SK) and the SeqLDA model, is proposed to analyze the texts of complaints.Research shows that LDA is suitable for the texts of automobile recalls, and SeqLDA is suitable for consumer complaints. The SK-SeqLDA model can significantly enhance the fitting effect of the sequential latent Dirichlet allocation model. In addition, body accessories and upholsteries are involved in many topics in the dataset of consumer complaints. A number of topics relate to the problems of braking in the automobile recalls dataset. The recalls of imported automobiles are more proactive, mainly focused on engines; the recalls of joint venture automobiles are primarily focused on vehicle controls and brake pedals, and the proportions of topics in self-owned brands are balanced. The results provide theoretical references for automobile manufacturers in management decisions, relevant government departments in defect investigation, and consumers in purchase and vehicle maintenance.

Key words: consumer complaints, automobile recalls, SK-SeqLDA model, nonparametric bootstrap test

中图分类号: