A Deep Learning Framework for Heterogeneous Spatial Social Networks: Modeling Human Mobility and Beyond

Keywords: GeoAI, human mobility, heterogenous graph, spatial representation learning
Abstract Type: Paper Abstract

Abstract

Modeling spatial networks has been a central focus of geospatial analysis, with applications across fields such as transportation, urban planning, and epidemiology. More recently, the integration of spatial and social networks into spatial social networks has emerged as a promising approach, leveraging both spatial and social dimensions to better understand and predict human dynamics. However, existing modeling methods for these networks rely on traditional techniques, using hand-crafted features that demand extensive domain expertise. With the advancements in artificial intelligence, especially deep learning, representation learning has demonstrated significant advantages over conventional feature engineering. In this study, we propose a novel framework that incorporates a heterogeneous spatial social network and a tailored encoding method to enable deep learning-based modeling of human-place interactions. Experimental results demonstrate that our framework effectively captures latent features of spatial social networks, achieving high performance in human mobility prediction. We also discuss potential applications of this framework, including mobility profiling, human activity intensity prediction, and mobility anomaly detection, underscoring its versatility and impact.