As the "World Center of Tourism and Leisure," Macao's daily tourism demand forecasting is crucial to the sustainable development of the tourism industry. Extracting valuable insights from daily tourism demand forecasting can assist decision-making by providing useful information to policymakers, businesses, and stakeholders in the tourism industry. The thesis aims to explore the use of deep learning (DL) models to enhance daily tourism demand forecasting in Macao by highlighting the challenges associated with daily tourism demand data, as it is known to contain various data features that are difficult to process using the traditional model. The thesis focuses on two separate essays that investigate different deep learning models for this purpose. The first essay examines the application of a specific deep learning model, integration of Variational Mode Decomposition (VMD) with Convolutional Neural Network (CNN) models, for daily tourism demand forecasting in Macao. Compared with the direct use of the deep learning model, which is widely applied in marketing literature, the proposed combination of the decomposition method (VMD) provides a better result in terms of forecasting accuracy. The second essay focuses on another deep learning model, the use of multiple large language models (LLMs) to estimate marketing concepts from big data and then integrate these concepts into traditional forecasting models. The essay highlights the advantages of multiple LLMs in capturing different aspects of user-generated review comments. It discusses the methodology of estimating satisfaction scores using LLMs. The evaluation of the forecast accuracy is achieved by integrating satisfaction scores estimated by multiple LLMs into traditional forecasting models. The findings from both essays provide theoretical and practical implications for daily tourism demand forecasting. The integration of VMD with CNN demonstrates improved forecasting accuracy by considering the multiscale data feature in the tourist arrival data. The utilization of multiple LLMs for estimating satisfaction scores enhances forecast accuracy by leveraging the capabilities of these models to analyze large text datasets. The linkage between satisfaction and tourism demand has been proposed for a long time in the literature using survey data and, for the first time, tested in empirical studies. The second essay highlights the possibility of utilizing LLMs to improve forecasting accuracy.