NARA: Anchor-Conditioned Relation-Aware Contextualization of Heterogeneous Geoentities

2026-05-12

AI Summary

Existing approaches struggle to jointly model the semantics, geometry, and multidimensional spatial relationships—such as metric and topological relations—of heterogeneous vector-based geographic entities like points, lines, and polygons, thereby limiting contextual representation capabilities. To address this challenge, this work proposes NARA, a self-supervised framework that, for the first time, integrates multiple types of spatial relations within a unified architecture. NARA leverages an anchor-conditioning mechanism and relation-aware contextual encoding to enable joint semantic–geometric–relational representation learning for heterogeneous geographic entities. Experimental results demonstrate that NARA significantly outperforms state-of-the-art methods across diverse tasks, including building function classification, traffic speed prediction, and point-of-interest recommendation, confirming the effectiveness and generalizability of its unified spatial relation modeling.

Abstract

Geospatial foundation models have primarily focused on raster data such as satellite imagery, where self-supervised learning has been widely studied. Vector geospatial data instead represent the world as discrete geoentities with explicit geometry, semantics, and structured spatial relations, including metric proximity and topological relationships. These relations jointly determine how entities interact within space, yet existing representation learning methods remain fragmented, often restricted to specific geometry types or partial spatial relations, limiting their ability to capture unified spatial context across heterogeneous geoentities. We propose NARA (Neural Anchor-conditioned Relation-Aware representation learning), a self-supervised framework for vector geoentities. NARA learns context-dependent representations by jointly modeling semantics, geometry, and spatial relations within a unified framework and captures relational spatial structure beyond proximity alone, enabling rich contextualized representations across heterogeneous geoentities of points, polylines, and polygons. Evaluation on building function classification, traffic speed prediction, and next point-of-interest recommendation shows consistent improvements over prior methods, highlighting the benefit of unified relational modeling for vector geospatial data.

Problem

• vector geospatial data
• heterogeneous geoentities
• spatial relations
• representation learning
• geospatial context

Innovation

• relation-aware
• vector geospatial data
• self-supervised learning
• heterogeneous geoentities
• spatial context