Tourism demand forecasting using social media data: A deep learning–based ensemble model with social media communication conversion rates

Social media has become a vital external driver of tourism demand, yet prior studies have largely overlooked its dynamic, nonlinear effects and the mechanisms by which online information shapes offline behavior. Using data from blogs, image-and-text platforms, and short-video platforms, this study applies transfer entropy to construct platform-specific Social Media Communication Conversion Indices that capture the directional and lagged causal influence of digital content on tourist behavior. To model the complex temporal dynamics of these indices, we propose a deep learning–based ensemble forecasting framework. Empirical results based on daily city- and attraction-level demand data show that our model consistently outperforms seven benchmark methods. This study advances theoretical understanding and offers a robust methodological foundation for tourism forecasting.