Understanding customer complaints from negative online hotel reviews: A BERT-based deep learning approach

Abstract

This paper utilizes the deep learning model based on BERT-BiLSTM-CRF in combination with the econometric model to examine how hotel customers’ complaints toward diverse service attributes contribute to their overall satisfaction. With our model, seven types of customer complaints, including service, facility, cleanliness, price, location, dining, and noise, can be automatically identified from hotel online reviews, achieving an F1 of 0.82 and a recall of 0.85. Econometrics analyses show that different types of complaints have varying degrees of impact on customer satisfaction. For example, in the hotel industry, service complaints show a stronger negative effect than cleanliness complaints, facility complaints, etc. Furthermore, the results of the robustness check show that our conclusions are consistent before and after COVID-19. Our findings contribute to the customer dissatisfaction literature and offer practical implications for service failure management in online travel platforms.