Unlocking travel narratives: a fusion of stacking ensemble deep learning and neural topic modeling for enhanced tourism comment analysis

User-generated comments are crucial in the domain of hotel bookings, especially in the fast-changing online planning and booking industry. Our research presents a sophisticated system that collects and organizes feedback from Booking’s website, with a specific emphasis on tourist cities in Morocco. The methodology utilizes XLNet for analyzing Arabic and French languages. It incorporates four advanced deep learning models, including LSTM, CNN, GRU, and BiLSTM. These models are merged using a stacking ensemble technique and a neural topic model to extract significant themes from reviews effectively. This novel strategy demonstrates a significant 0.975 increase in accuracy compared to traditional methods, such as GPT and BERT. In addition, the study methodically investigates the influence of dataset scaling by comparing different levels of dataset scaling.