One Capital, Many Provinces: Spatial Concentration and the Collaboration Network of Iranian Artificial Intelligence Research, 2010–2021
- 1Shahid Beheshti University, Tehran
Abstract
Purpose. Research capacity tends to concentrate in capital cities, which can limit the diffusion of a fast-moving field such as artificial intelligence (AI). We measure how concentrated Iranian AI research is in Tehran, whether it has become less so, how institutions across cities collaborate, and whether work outside Tehran is less visible, benchmarked against Türkiye and Egypt. Design and methods. Using OpenAlex records of AI articles published in 2010–2021 (before a 2022 coverage break), we located institutions by city and computed concentration indices (largest-city share, Herfindahl–Hirschman index, Gini and Theil decomposition) for Iran, Türkiye and Egypt with bootstrap intervals; analysed 9,353 Iranian papers by location of their institutions (year trend, citation visibility with year, topic and team-size controls); and built the domestic institution co-authorship network (143 institutions, 4,522 tie weights), testing within-city ties by permutation. National multi-campus systems that OpenAlex assigns to Tehran (Islamic Azad University as a whole and Payame Noor University) are excluded and reported as a sensitivity case. Findings. Tehran institutions hold 46.4% of Iranian AI institution–paper slots (95% CI 29.4–61.0), against 26.2% for Istanbul (18.7–37.7) and 31.0% for Cairo (17.2–48.4; 44.3% for Greater Cairo); the city HHI is 0.229 versus 0.110 and 0.134, and the share rises to 51.6% if multi-campus entities are included. Concentration declined: the share of papers with a Tehran institution fell from 55.9% in 2010 to 44.1% in 2021 (−1.0 percentage points per year, p < 0.001). The network is Tehran-centred: 72.9% of tie weight involves a Tehran institution, 61.7% of inter-city ties, and 95.2% of the 105 non-Tehran institutions have at least one tie to Tehran; 45.4% of ties are within-city against 32.3% expected by chance (permutation p = 0.002). Visibility differences are small: 28.8% of Tehran-only papers were in the citation top-10%, against 26.7% for other-city papers (adjusted gap −2.8 pp, 95% CI −4.8 to −0.9) and 27.2% for papers combining Tehran and other cities (−5.3 pp, −8.4 to −2.2). Originality. This is, to our knowledge, the first quantification of the geography and collaboration network of Iranian AI research benchmarked against comparable systems, and it shows how national multi-campus entities can bias geographic analyses built on affiliation data.
Related
- Human–artificial intelligence collaboration in emergency medical services: from diagnostic algorithms to shared decision support
- Concentration Determination of Aspirin Deposit Patterns Using Quantitative Image Analysis and Artificial Intelligence
- Polymath Collaboration and AI Accelerate Bounded Prime Gap Breakthroughs — E8 Intelligence Research
- Polymath Collaboration and AI Advance Prime Gap Bounds — E8 Intelligence Research
- Polymath and AI Collaboration Narrows Prime Gap Bounds to 246 — E8 Intelligence Research
BibTeX
@misc{ghorbani2026capital,
title = {{One Capital, Many Provinces: Spatial Concentration and the Collaboration Network of Iranian Artificial Intelligence Research, 2010–2021}},
author = {Pooria Ghorbani},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23448574},
url = {https://zenodo.org/records/23448574},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - One Capital, Many Provinces: Spatial Concentration and the Collaboration Network of Iranian Artificial Intelligence Research, 2010–2021
AU - Pooria Ghorbani
PY - 2026
DA - 2026/10/18
PB - Zenodo
DO - 10.5281/zenodo.23448574
UR - https://zenodo.org/records/23448574
AB - Purpose. Research capacity tends to concentrate in capital cities, which can limit the diffusion of a fast-moving field such as artificial intelligence (AI). We measure how concentrated Iranian AI research is in Tehran, whether it has become less so, how institutions across cities collaborate, and whether work outside Tehran is less visible, benchmarked against Türkiye and Egypt. Design and methods. Using OpenAlex records of AI articles published in 2010–2021 (before a 2022 coverage break), we located institutions by city and computed concentration indices (largest-city share, Herfindahl–Hirschman index, Gini and Theil decomposition) for Iran, Türkiye and Egypt with bootstrap intervals; analysed 9,353 Iranian papers by location of their institutions (year trend, citation visibility with year, topic and team-size controls); and built the domestic institution co-authorship network (143 institutions, 4,522 tie weights), testing within-city ties by permutation. National multi-campus systems that OpenAlex assigns to Tehran (Islamic Azad University as a whole and Payame Noor University) are excluded and reported as a sensitivity case. Findings. Tehran institutions hold 46.4% of Iranian AI institution–paper slots (95% CI 29.4–61.0), against 26.2% for Istanbul (18.7–37.7) and 31.0% for Cairo (17.2–48.4; 44.3% for Greater Cairo); the city HHI is 0.229 versus 0.110 and 0.134, and the share rises to 51.6% if multi-campus entities are included. Concentration declined: the share of papers with a Tehran institution fell from 55.9% in 2010 to 44.1% in 2021 (−1.0 percentage points per year, p < 0.001). The network is Tehran-centred: 72.9% of tie weight involves a Tehran institution, 61.7% of inter-city ties, and 95.2% of the 105 non-Tehran institutions have at least one tie to Tehran; 45.4% of ties are within-city against 32.3% expected by chance (permutation p = 0.002). Visibility differences are small: 28.8% of Tehran-only papers were in the citation top-10%, against 26.7% for other-city papers (adjusted gap −2.8 pp, 95% CI −4.8 to −0.9) and 27.2% for papers combining Tehran and other cities (−5.3 pp, −8.4 to −2.2). Originality. This is, to our knowledge, the first quantification of the geography and collaboration network of Iranian AI research benchmarked against comparable systems, and it shows how national multi-campus entities can bias geographic analyses built on affiliation data.
ER -