Forecasting fine-grained tourism demand with peripheral surroundings: implications for destination networking

Fine-grained, near-real-time forecasting is essential for destination management. Recent AI-based models have improved tourism forecasting by examining spatial interactions among attractions, but they overlook external factors like surroundings, which are crucial in shaping tourists’ decisions to visit attractions. This study uses the Attribute-Augmented Spatiotemporal Graph Convolutional Network model to capture surrounding and environmental features. Forecasting experiments for 134 attractions in Beijing, China, validated the model’s efficacy. Results show that including surroundings significantly enhances forecasting performance. Moreover, surrounding features exhibit stronger forecasting power compared to environment features. Valuable insights for advances tourism demand theory and implications for destination networking are provided.