Forecasting daily tourism demand with multiple factors

Abstract

Various factors have contributed to forecasting tourism demand. Although deep learning methods can achieve accurate results, they haven't considered the temporal heterogeneity of multiple factors and lack interpretability. This study proposes a novel deep learning method for daily tourism demand forecasting. Benefiting from the encoder-decoder architecture, our method adequately exploits the temporal heterogeneity of multiple factors. Based on the attentional mechanism, our method provides an interpretation of tourism demand from both factors and temporal persistence patterns. The effectiveness of our method is verified through an empirical study of two tourist attractions before and during COVID-19. Our method compensates for the uninterpretability of deep learning models, which allows tourism managers to obtain deeper insights.