Search papers, labs, and topics across Lattice.
This paper introduces GeoFlow, a framework that enhances origin-destination (OD) flow modeling by integrating geospatial attributes and employing a specialized geometric-intrinsic fusion encoder. By combining graph attention mechanisms with coordinate-aware encoders and an axial-global attention decoder, GeoFlow effectively captures long-range dependencies and competitive interactions in mobility data. Empirical results demonstrate that GeoFlow significantly outperforms existing methods in predictive accuracy and generative fidelity, offering a more nuanced understanding of urban mobility patterns.
GeoFlow reveals that integrating geospatial attributes can dramatically enhance the accuracy and diversity of urban mobility predictions.
Origin-destination (OD) flow modeling underpins urban planning and mobility analysis, but prevailing graph-based methods often neglect salient geographic attributes, limiting their ability to model long-range and multi-area dependencies. In this paper, we introduce GeoFlow, a novel framework that (i) augments area representations with geospatial attributes, including relative positions, k-hop and geodesic distances, (ii) employs a specialized geometric-intrinsic fusion encoder design that combines graph attention for intrinsic area signals with coordinate-aware encoders for global structure, and (iii) adopts an axial-global attention decoder to capture OD-specific competitive dependencies. For OD flow generation, GeoFlow is paired with flow matching models to produce more authentic and diverse mobility samples. Empirically, GeoFlow achieves superior performance in predictive accuracy, while substantially improving generative fidelity and diversity. Ablation and analytical studies confirm the contribution of each component. Code is available at https://github.com/ZheruiHuang/GeoFlow.
