Data‐driven inference of interactions among multiple tourist attractions for hourly demand forecasting

Abstract

AbstractExisting tourism demand forecasting models mainly focus on forecasting demands of relatively long time spans at a single destination. These studies lack considering the evolution of demand patterns or presume fixed interaction structures among multiple destinations, limiting their applications during uncertain times when demands and their interactions could evolve fast. This study simultaneously infers interactions among multiple tourist attractions from tourist arrival data and learns temporal patterns for forecasting short‐term tourist arrivals at these attractions. We achieve this goal by working on data of high spatiotemporal resolution and developing a variational autoencoder framework, where a data‐driven encoder infers the interactions and a decoder learns the short‐term dynamics of tourism demand.