Adaptive shopping malls: Probabilistic spatial layouts with graph neural networks

We propose an adaptive method for optimizing shopping mall layouts via graph neural networks (GNNs). Traditional architectural datasets, often presented as floor plans rather than graphs, explore the relationship between mall layouts and pedestrian flow through static spatial structure analysis. However, these methods rely on specific environments and manual interventions, making automation and dynamic prediction challenging and limiting their applicability across diverse scenarios. To address this, the study constructs graph datasets by automatically extracting key pathways and shop layouts from mall floor plans. A GNN-based model is then developed to predict dynamic pedestrian flows, learn spatial node relationships and optimize layouts in real time via a gravity model. This flexible approach, which is applicable to various architectural and urban design scenarios, supports data-driven spatial optimization, enhancing design adaptability and efficiency.