Abstract: With the rapid development of the Internet, online complaints regarding automobile quality have emerged as an efficient and convenient means for automotive enterprises to gather user feedback. Consequently, this trend has necessitated the exploration of effective methods to leverage these online complaints in order to mine and monitor issues pertaining to automobile quality. Given that the frequency of online complaints is influenced by numerous factors, including automobile quality, sales, the development of the Internet, and public awareness of rights protection, it exhibits significant autocorrelation and dynamics. However, the existing methods, which are often premised on the assumption that the in-control processes remain unchanged and adhere to models with fixed parameters and constant control limits, have proven to be unreliable in practical applications for delivering dependable monitoring and timely warnings.
In this context, the motivation behind our research stems from the urgent need for automotive enterprises to effectively harness the wealth of data contained in online complaints. The problem we aim to address is how to develop an adaptive monitoring system that can dynamically capture and analyze these complaints, accurately identifying trends and potential quality issues in real-time. By doing so, enterprises can take proactive measures to address concerns, enhance product quality, and ultimately improve customer satisfaction.
To this end, we propose a dynamic method for automobile quality monitoring, denoted as the SDINAR chart, based on the sliding window Integer-valued Autoregressive (INAR) model. Firstly, we incorporate the sliding window technique, a concept that allows us to analyze data in segments over time. The maximum entropy principle is used to select the optimal window period, ensuring that we capture the most informative and representative data slices. Within each window, an INAR model is fitted to the data, and the model parameters are obtained through conditional maximum likelihood estimation. As the sliding window moves forward, the model parameters are adjusted accordingly, resulting in the establishment of a variable coefficient INAR model, which realizes the dynamic fitting of the automobile quality online complaint data. Subsequently, the conditional probability of the one-step-ahead prediction residuals is calculated, and given a specific confidence level, a confidence interval for the residuals is obtained through Monte-Carlo simulation. This leads to the development of a time-varying control limit monitoring method. This method enables us to set dynamic thresholds that can trigger alerts when the quality of automobile complaints deviates from expected norms, thereby facilitating prompt and informed decision-making.
We select “Volkswagen Sagitar” as the subject of our case study, a car brand that has consistently maintained high sales figures in recent years. To gather relevant data, we source complaint texts pertaining to Volkswagen Sagitar from “Automobile Quality Website” (http://www.12365auto.com), a national platform dedicated to receiving and addressing consumer complaints related to automobiles. Utilizing the proposed SDINAR chart, we conduct an in-depth and dynamic analysis and monitoring of the complaint data concerning the defect topic of “abnormal brake noise” from January 2015 to June 2023. To evaluate the efficacy of the proposed method, we compare its performance with traditional monitoring approaches. The results of this comparison underscore the effectiveness of the SDINAR chart in dynamic monitoring and early warning.
This research not only provides automotive enterprises with a novel tool to evaluate the dynamic state of automobile quality, but also supports decision-making based on monitoring and alert feedback. However, there are still opportunities for further exploration. In future studies, we plan to investigate the correlations between multiple complaint topics and establish a network model to comprehensively monitor the overall state of automobile quality. Additionally, we aim to incorporate additional monitoring indicators beyond complaint frequency, such as the time interval between complaints and textual features extracted from the complaints, to develop a more efficient monitoring chart from various perspectives. We believe that these enhancements will further strengthen the effectiveness and comprehensiveness of our approach in monitoring and assessing automobile quality.

Key words: automobile quality, online complaints, sliding window, INAR, dynamic control chart

摘要： 随着互联网普及率的逐年提升,汽车质量网络投诉为汽车企业获取用户反馈提供了一条高效、便捷的途径,这也对如何有效利用在线投诉挖掘和监控汽车质量问题提出了要求。由于网络投诉频数受到汽车质量、销售量、互联网发展、大众维权意识等诸多因素的影响,呈现出显著的自相关性和动态性,采用现有的固定参数和常数控制限监控方法不再可靠。为此,本文提出了一种基于滑动窗口整值自回归(INAR)模型的汽车质量动态监控方法。应用最大熵原理选取最优窗口期数,基于滑动窗口建立变系数INAR模型,实现对汽车质量网络投诉数据的动态拟合。应用向前一步预测残差的置信区间,确定不同时点的可变控制限,实现对于汽车质量的动态监测与预警。最后以大众速腾汽车为例,应用本文提出的方法,对其在车质网中2015年1月至2023年6月关于“刹车异响”缺陷主题下的投诉数据进行了动态分析和监控,并给出了两种方法的监测效果比较,说明了本文方法的有效性。车企可以依据监测预警反馈评估汽车质量的动态状况并进行决策。

关键词: 汽车质量, 网络投诉, 滑动窗口, INAR, 动态控制图