Comparing World City Networks by Language: A Complex-Network Approach

Wenjia Zhang, Jiancheng Zhu, Pu Zhao

TL;DR

This paper explores how language shapes our understanding of global city networks (WCNs). By analyzing city co-occurrences on webpages in English, German, and French, they reveal that different languages highlight distinct network structures and city roles, suggesting that a singular view of globalization is insufficient.

What Problem Does This Paper Address?

The concept of "world cities" and their interconnectedness (World City Networks or WCNs) is a key lens for understanding globalization. However, most WCN studies tend to focus on economic ties and often assume a universal, singular model of globalization. This overlooks the potential for "multiple globalizations" influenced by cultural and linguistic differences. Specifically, the paper addresses the gap in research that compares WCNs across different languages and cultures, arguing that our perception of global hierarchies and connections is likely shaped by the language we use. Existing methods for analyzing WCN structures, like community detection, can be biased by pre-defined assumptions.

Methods

The authors employ a complex-network approach to analyze WCNs. They use webometrics to collect data on city linkages, defining the "weight" of a connection between two cities as the number of webpages where both city names appear. This data is gathered using the Advanced Google Search engine across three languages: English, German, and French.

For network analysis, they move beyond traditional macroscale (city rankings, overall network properties) and microscale (individual city centrality) analyses to focus on mesoscale structures. To do this, they utilize a weighted stochastic block model (WSBM). This method is novel because it doesn't pre-define the network's structure (like core-periphery or community structures) but rather infers latent grouping patterns probabilistically. This allows for the detection of more nuanced and potentially hybrid structures. They also introduce an Incremental Fit Index (IFI) to quantitatively compare the similarity between different WCN partitions (e.g., language-specific structures vs. continent-based structures).

Datasets

The study uses a large dataset of intercity linkages derived from web crawling. Specifically, they collected co-occurrence data from webpages in English, German, and French. The data was gathered between May and June 2016, involving approximately 0.138 trillion webpages searched in total. The identified world cities (126 in total) were selected based on their frequent appearance in influential WCN studies over the past 30 years.

Key Insights and Ideas

  • Language as a Lens for Globalization: The study powerfully demonstrates that language isn't just a communication tool but actively shapes how we perceive and construct global city networks. Different languages highlight different city roles and connections.
  • Beyond a Singular Globalization: The findings challenge the idea of a monolithic "world city network." Instead, they support the concept of "multiple globalizations," where cultural and linguistic contexts create distinct network realities.
  • WSBM's Advantage: The WSBM proves to be a valuable tool for uncovering mesoscale structures without pre-conceived notions, revealing more complex and potentially hybrid network formations than traditional methods.
  • Territoriality vs. Global Reach: English-language WCNs appear more "globalized," with connections spanning continents. In contrast, German and French WCNs show more "territorial" tendencies, with stronger regional concentrations.
  • Cultural Proximity Matters: Cities with shared colonial histories or immigrant populations tend to cluster together, even if geographically distant, suggesting that cultural ties play a significant role in virtual connectivity.

Key Results

  • Macroscale Differences: While New York consistently ranks as the top city across all three languages, the rankings of other cities and the distribution of connections vary significantly. The English WCN shows a more even distribution of centralities globally, while German and French WCNs are more concentrated in Europe.
  • Mesoscale Structures:
    • The English WCN exhibits a "multicores-periphery" structure, with a global core (New York) and regional cores.
  • The German WCN shows a "duo-cores-periphery" structure, with New York as a global core and a regional core (largely German cities).
  • The French WCN presents a more typical "core-periphery" structure.
  • IFI Values: The study found that the mesoscale structure of the English WCN has a higher "generalization ability" for interpreting other language WCNs, suggesting it might represent a more generalized global imagination. However, using the English WCN structure to understand German or French WCNs can lead to significant deviations (around 40%).

Limitations and Future Work

  • Computational Burden: The WSBM approach, while powerful, can be computationally intensive, especially for very large networks.
  • Webpage Content Nuance: The current method of using co-occurrence frequency might not fully capture the specific nature of intercity linkages. Future work could improve this by analyzing the type of content on webpages (e.g., business news, travel information).
  • Language Coverage: The study focused on English, German, and French. Expanding to other major languages, particularly Chinese and Japanese, is crucial given their growing global importance.
  • Non-Economic Linkages: While this study moves beyond purely economic WCNs, further exploration of non-economic linkages across languages would be beneficial.
  • Formation of Mesoscale Structures: Understanding the underlying reasons for the formation of these observed mesoscale structures remains an important area for future research.

Summary updated

Abstract

City networks are multiplex and diverse rather than being regarded as part of a single universal model that is valid worldwide. This study contributes to the debate on multiple globalizations by distinguishing multiscale structures of world city networks (WCNs) reflected in the Internet webpage content in English, German, and French. Using big data sets from web crawling, we adopted a complex-network approach with both macroscale and mesoscale analyses to compare global and grouping properties in varying WCNs, by using novel methods such as the weighted stochastic block model (WSBM). The results suggest that at the macro scale, the rankings of city centralities vary across languages due to the uneven geographic distribution of languages and the variant levels of globalization of cities perceived in different languages. At the meso scale, the WSBMs infer different grouping patterns in the WCNs by language, and the specific roles of many world cities vary with language. The probability-based comparative analyses reveal that the English WCN looks more globalized, while the French and German worlds appear more territorial. Using the mesoscale structure detected in the English WCN to comprehend the city networks in other languages may be biased. These findings demonstrate the importance of scrutinizing multiplex WCNs in different cultures and languages as well as discussing mesoscale structures in comparative WCN studies.
BibTeX
@article{zhang2021comparing,
  author = {Wenjia Zhang and Jiancheng Zhu and Pu Zhao},
  title = {{Comparing World City Networks by Language: A Complex-Network Approach}},
  journal = {Isprs Int. J. geo inf},
  year = {2021},
  doi = {10.3390/ijgi10040219},
}
Figure