Tourism demand forecasting using graph neural network

Abstract

Forecasting tourism demand is pivotal for crafting policies that nurture sustainable tourism. Despite its significance, the field grapples with challenges that limit its wider application, particularly in multi-step forecasting among diverse tourist attractions across scenarios. The absence of customised approaches in this field sparked our initiative to develop a cutting-edge graph neural network, specifically designed to address these needs. At the heart of the proposed algorithm lies a belief: graph neural network’s capability to analyze spatiotemporal correlations, fused with a deep learning architecture, makes it uniquely equipped to address the complexities of demand forecasting. This study develops three modules: dual correlation matrix, multi-head coordinate attention, and automatic control. These are meticulously crafted and synergized to tackle the methodological challenges prevalent in tourism demand forecasting. The experimental findings demonstrate proposed forecasting algorithm surpasses existing state-of-the-art algorithms in trials involving data from three renowned Chinese tourist cities. Through the validity of the algorithm, the conclusion supports the policy implication in developing multi-dimensional sustainable tourism, integrating the insights into future demand trends. This study not only advances the theoretical understanding of sustainable tourism and demand forecasting but also marks a significant stride in the intersection of artificial intelligence and tourism management.

Topics & Concepts