May, 2024

学习具有异构图的地理空间区域嵌入

TL;DRGeospatial embeddings are crucial for city analytics and earth monitoring, but learning comprehensive region representations is challenging; this paper introduces GeoHG, an effective heterogeneous graph structure, to learn region embeddings by addressing intra-region feature representation and inter-region dependencies, demonstrating superior performance in geo-prediction tasks even with limited training data.