Identifying urban land use through higher-order spatial interactions

Crowd flow connects various geographic spaces in cities, revealing inter-regional associations that are crucial for urban land – use identification. Existing research mainly focuses on binary connections between pairs of regions, overlooking associations among multiple regions. Addressing this gap, we propose a network model that uses a hypergraph neural network to extract key features of higher-order connections for urban land – use classification. Additionally, a similarity enhancement module is incorporated to augment the recognition capabilities of the model. Compared with graph neural networks, incorporating higher-order connections among multiple regions improves urban land – use identification. Metrics show a decrease of 0.4 in L1 distance, 2.35 in KL divergence, and 0.14 in Chebyshev distance, while cosine similarity increased by 0.25, particularly in areas with high crowd mobility. The similarity enhancement module further refines the ability of the model to capture regional similarities, particularly effective in large contiguous areas or regions with extreme points of interest distributions. Additionally, the degree of land – use mixing and inter-regional movement influences the effectiveness of higher-order connections in recognizing land use, with noticeable negative and positive impacts, respectively. This study provides methods and insights for the utilization of higher-order associations in land-use identification and urban studies on crowd flow.