Learning Effective Road Network Representation with Hierarchical Graph Neural Networks

Abstract

Road network is the core component of urban transportation, and it is widely useful in various traffic-related systems and applications. Due to its important role, it is essential to develop general, effective, and robust road network representation models. Although several efforts have been made in this direction, they cannot fully capture the complex characteristics of road networks. In this paper, we propose a novel Hierarchical Road Network Representation model, named HRNR, by constructing a three-level neural architecture, corresponding to "functional zone", "structural regions" and "road segments", respectively. To associate the three kinds of nodes, we introduce two matrices consisting of probability distributions for modeling segment-to-region assignment or region-to-zone assignment. Based on the two assignment matrices, we carefully devise two reconstruction tasks, either based on network structure or human moving patterns. In this way, our node presentations are able to capture both structural and functional characteristics. Finally, we design a three-level hierarchical update mechanism for learning the node embeddings through the entire network. Extensive experiment results on three real-world datasets for four tasks have shown the effectiveness of the proposed model.