Analysis of the Topological Similarity of Street Networks in Cities Around the World Using Graph2vec

Abstract

This study employs graph2vec, a graph-based machine learning method, to quantify the topological characteristics of street networks in 2,450 cities worldwide. The results reveal distinct regional differences, particularly among Asia, Europe, and America, and show that Asia exhibits notable internal diversity, with subregions displaying unique structural features. These patterns suggest that historical context and infrastructure development have shaped urban form. The findings underscore the limitations of directly applying insights from European and American cities to Asia and highlight the need for region-specific analyses. While this study focuses on vehicle-accessible networks, future work should incorporate multimodal transport systems, facility distributions, and topographic conditions to better capture urban morphology. Additionally, improving the interpretability of embedding techniques and comparing alternative graph learning methods are essential next steps for advancing applications in urban and transport planning.