Analyzing world city network by graph convolutional networks.

Area of Science:

• Urban Studies
• Network Science
• Data Science

Background:

• The World City Network (WCN) is crucial for understanding global economic hierarchies.
• Existing WCN construction methods, like Peter Taylor's, suffer from data deficiencies and update delays.
• A real-time, data-sufficient approach is needed to accurately represent the dynamic WCN.

Purpose of the Study:

• To construct an updated World City Network using novel data sources.
• To introduce and evaluate a new ranking algorithm, GCNRank, for world cities.
• To address data insufficiency and update delays in existing WCN models.

Main Methods:

• Utilized publicly available, real-time global flight route data to build the WCN.
• Employed advanced Graph Convolutional Networks (GCNs) for network analysis.
• Introduced GCNRank, a novel centrality measure for city ranking within the WCN.

Main Results:

• Successfully constructed a real-time World City Network, overcoming data limitations.
• GCNRank demonstrated superior performance in representing city rankings.
• The new method effectively avoids local optima issues present in traditional centrality measures.

Conclusions:

• Real-time flight data provides a robust foundation for constructing dynamic World City Networks.
• GCNRank offers a more granular and accurate assessment of global city importance.
• This approach enhances our understanding of contemporary global urban systems and their interconnections.