Hotel Demand Forecasting via Booking’s Comments Using Sentiment Analysis and Topic Modeling Techniques

Planning and booking are one of the major commercial uses of the Internet nowadays. With the rise of Web 2.0, user-generated comments, reviews, and trip reports are becoming more important sources of information. Such reviews are crucial when reserving a hotel because they are more precise and exhaustive, unaffected by marketing variables such as hotel homepages or catalog descriptions representing authentic guest experiences. Consequently, hotel administration can use these reviews to maintain quality control. We present a system that collects French comments from the Booking website about Morocco’s tourist cities. This system generates classified overviews of such comments and discovers topics in each class (positive and negative) to identify actionable insights from the reviews and assist the hotel in enhancing guest satisfaction. In the suggested method, we first used XLNet for French languages to obtain word embeddings from the gathered dataset; next, we built four deep learning algorithms, CNN, GRU, LSTM, and BiLSTM that take the word embeddings as input. The outputs of these algorithms are then combined using the stacking ensemble model, with a Multilayer perceptron functioning as a meta learner. After acquiring the two classes, an LDA topic model was developed to extract themes from each. Applying the proposed method to three French datasets indicates that our approach improved accuracy by 0,973 compared to basic deep learning methods employing various contextualized word embedding models, specifically GPT and BERT.