Daily tourism demand forecasting and tourists' search behavior analysis: a deep learning approach

During trip planning and booking, tourists can search the web using different devices, such as personal computers (PC) or mobile devices. Search engine data can help understand the diverse aspects of tourists' preference and search behavior. Through the integration of Baidu Index PC search (PCS) and mobile phone search (MPS) volume on a tourism attraction in China, this exploratory study constructed the state-of-the-art Long Short-Term Memory (LSTM) deep learning models to investigate the influence of different search volume data and time step on the optimization of model parameters and forecast accuracy, and capture tourists' search behaviors using different devices. The models were evaluated with mean absolute error (MAE), root mean square error (RMSE), symmetric mean absolute percentage error (SMAPE). Results indicated that including Baidu search volume data can improve the daily forecasting performance of LSTM models (RMSE by 13.51%, MAE by 16.1%, and SMAPE by 11.04%). The LSTM model with the optimal subsets of MPS volumes achieved the best performance (RMSE = 247.91, MAE = 187.27, SMAPE = 0.4453). Tourists' search behaviors using different devices vary. More search query data does not always lead to higher forecasting accuracy. Web big data, especially MPS volumes, should be incorporated into short-term daily tourism demand forecasting.