We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Physics > Physics and Society

Title: Quantifying spatial homogeneity of urban road networks via graph neural networks

Abstract: The spatial homogeneity of an urban road network (URN) measures whether each distinct component is analogous to the whole network and can serve as a quantitative manner bridging network structure and dynamics. However, given the complexity of cities, it is challenging to quantify spatial homogeneity simply based on conventional network statistics. In this work, we use Graph Neural Networks to model the 11,790 URN samples across 30 cities worldwide and use its predictability to define the spatial homogeneity. The proposed measurement can be viewed as a non-linear integration of multiple geometric properties, such as degree, betweenness, road network type, and a strong indicator of mixed socio-economic events, such as GDP and population growth. City clusters derived from transferring spatial homogeneity can be interpreted well by continental urbanization histories. We expect this novel metric supports various subsequent tasks in transportation, urban planning, and geography.
Comments: 22 pages, 5 figures
Subjects: Physics and Society (physics.soc-ph); Machine Learning (cs.LG); Social and Information Networks (cs.SI)
Cite as: arXiv:2101.00307 [physics.soc-ph]
  (or arXiv:2101.00307v1 [physics.soc-ph] for this version)

Submission history

From: Jiawei Xue [view email]
[v1] Fri, 1 Jan 2021 19:45:04 GMT (3020kb)

Link back to: arXiv, form interface, contact.