Data-driven quality management applying sentiment analysis and statistical process control: A Malaysian restaurant case study

Abstract

Despite the growing availability of customer feedback data, restaurant chains often face challenges in translating customer feedback into actionable quality improvements. To address this problem, this research aims to investigate operational process inefficiencies and food quality–related aspects of customer satisfaction. For this purpose, customer review data and operational activity data are collected from the central kitchen of a Malaysian restaurant chain. The data are first pre-processed through cleaning, transformation, and dimensionality reduction. Tokenization, stemming and part of speech tagging are applied to the customer review data. Subsequently, we utilize the BERT model, PyABSA model and the exponentially weighted moving average (EWMA) control chart to achieve the objectives. The performance of the BERT and PyABSA models is then compared. The sentiment analysis results are visualized in bar chart, Pareto chart and cause-and-effect diagram. The evaluation results indicate that the PyABSA model achieves higher accuracy and weighted F1 score compared to the BERT model. The findings suggest that most of the food quality complaints are related to duck, meat, and sauce products. Next, the EWMA control chart is analysed using metrics including the out-of-control percentage, in-control percentage, average run length, and identified out-of-control points. Based on the EWMA analysis, specific out-of-control dates are identified, highlighting the need for targeted strategies to improve operational processes during these periods. The results of this study provide valuable insight for the restaurant chain to improve operational processes and effectively address customer complaints related to food quality.