Customer emotion detection and analytics in hotel and tourism services using multi-label classificational models based on ensemble learning

Customer reviews play a crucial role in a company's success, particularly by amplifying the Electronic Word-of-Mouth (eWOM) effect. This study aims to harness ensemble learning-based multi-label classification models to analyze customer opinions, introducing a research model to identify customer emotions across four dimensions: happy, hopeful, depressed, and angry. We compiled an extensive dataset of 415,998 feedback records from tourists about hotels in Vietnam, sourced from reputable websites such as Agoda.com, Booking.com, and Traveloka.com. Our findings reveal that the stacking model demonstrated superior performance, achieving an Exact Match Ratio, Precision, Recall, and F1-score of 88%, 91%, 87%, and 89%, respectively. Using the same training dataset and preprocessing techniques, we conducted experiments to compare our method with two other models: Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). Although the recall scores range from 60 to 73%, the exact match ratio reached as high as 87%, indicating that our research model outperforms other methods. The model employed in this investigation was validated exclusively with select algorithms and multi-label classifiers, and it contributes a Vietnamese hotel reservation dataset for future reference. The study has not yet incorporated real-time data, and prediction accuracy may be influenced by the limited amount of data sourced from the web and the incompleteness of preprocessing procedures.