Options
Convolutional Neural Networks for Automated and Non Intrusive Measurement of Customer Satisfaction in Restaurants
Journal
Tourism and Hospitality
ISSN
2673-5768
Date Issued
2025-12-03
Author(s)
Abstract
Customer satisfaction (CS) is a cornerstone of competitiveness in the hospitality sector, particularly in restaurants, where service interactions are highly sensory and time-sensitive. Traditional measurement instruments, such as SERVQUAL, SERVPERF, and the American Customer Satisfaction Index, provide valuable diagnostic insights but remain limited by recall bias, social desirability, and delayed feedback. Advances in deep learning now enable non-intrusive, real-time monitoring of customer experience.
This study evaluates the feasibility of using a convolutional neural network (CNN) to automatically classify customer satisfaction based on facial expressions captured at the point of payment in a restaurant. From an initial dataset of over 5000 images, 2969 were validated and labeled through a binary self-report mechanism.
The CNN, implemented with transfer learning (MobileNetV2), achieved robust performance, with 93.5% accuracy, 92.8% recall, 91.0% F1-score, and an area under the ROC curve of 0.93. Comparative benchmarks with Support Vector Machine and Random Forest classifiers confirmed the superiority of the CNN across all metrics. The findings highlight CNNs as reliable and scalable tools for continuous CS monitoring, complementing rather than replacing classical survey-based approaches.
By integrating implicit, real-time signals with traditional instruments, restaurants can strengthen decision-making, enhance service quality, and co-create personalized experiences while addressing challenges of explainability, external validity, and data ethics.
This study evaluates the feasibility of using a convolutional neural network (CNN) to automatically classify customer satisfaction based on facial expressions captured at the point of payment in a restaurant. From an initial dataset of over 5000 images, 2969 were validated and labeled through a binary self-report mechanism.
The CNN, implemented with transfer learning (MobileNetV2), achieved robust performance, with 93.5% accuracy, 92.8% recall, 91.0% F1-score, and an area under the ROC curve of 0.93. Comparative benchmarks with Support Vector Machine and Random Forest classifiers confirmed the superiority of the CNN across all metrics. The findings highlight CNNs as reliable and scalable tools for continuous CS monitoring, complementing rather than replacing classical survey-based approaches.
By integrating implicit, real-time signals with traditional instruments, restaurants can strengthen decision-making, enhance service quality, and co-create personalized experiences while addressing challenges of explainability, external validity, and data ethics.