A multi-scale network analysis of international research collaboration

Abstract

International research collaboration networks are commonly analyzed using aggregate indicators that summarize national participation across all forms of partnership. However, such measures can obscure structural differences across collaboration scales, from bilateral to multi-partner engagements. Here, we show that national collaboration profiles are systematically scale-dependent, with countries occupying different structural positions in bilateral and multi-partner collaborations. Analyzing 1.4 million computer science journal articles (2000–2023), we stratify co-authorship networks into scale-specific subnetworks and compute structural indicators capturing centrality, cohesion, and partner diversity at each scale. Using dimensionality reduction (UMAP) and density-based clustering (HDBSCAN), we identify recurring multi-scale collaboration profiles that emerge across all seven computer science subfields. These profiles distinguish structurally central, intermediate, scale-selective, and peripheral forms of participation that are not readily observable in aggregate collaboration measures. Comparison with the Global Innovation Index (GII) shows that structurally central profiles are generally associated with stronger GII standings, while also revealing cases where network-based structural positioning is not reflected in composite indicators. Temporal analysis across three periods (2000–2007, 2008–2015, 2016–2023) further indicates a transition from coarse core–periphery structures toward increasingly differentiated collaboration profiles, with multiple countries shifting between profiles over time. Together, these findings extend informetric approaches to international collaboration by introducing a scale-sensitive framework for analyzing national positioning within journal-based computer science collaboration networks.

The authors' abstract, as published at the source. Journal of Informetrics, 2026 · DOI ↗