Mapping Scientific Frontiers: Network Embeddings Reveal Hidden Structures in Global Research Mobility

• Akhil Veluru

  University of Texas at Dallas

  Author

The linkage of research organisations going far beyond geography holds significance in understanding global scientific mobility and in developing a better communication mechanism; the need for an analytical framework is felt. This study introduces a representation learning method in which researchers’ co-affiliation trajectories are treated as sentence-like sequences. The authors employ a skip-gram with negative sampling (SGNS) model that embeds institutions into a high-dimensional space. In contrast to traditional approaches that use geographical distance, our learned embeddings capture the multi-dimensional cognitive proximity, organisational proximity, cultural proximity and linguistic proximity we find in the actual movements of millions of researchers. Evidence suggests that the cosine similarity between institutional embeddings can explain more than two times the variance in observed researcher flows as compared to distance. When incorporated into an augmented gravity model, the embedding-based predictor significantly outperforms in predictive accuracy for both intra-national and international mobility, achieving Pearson correlations of 0.79 and 0.76, respectively, vs 0.54 and 0.49 for geographic distance. The UMAP mapping visualisation helps to interpret the clusters corresponding to language groups, historical legacies, local academic ecosystems, eg, co-location of French-speaking institutions across continents or co-location at the level of states in the US. Network embeddings can successfully recover the hidden structure underlying international scientific mobility. Network embeddings, therefore, in development, provide a new and powerful data-driven tool for science policy and research on knowledge diffusion. 

• PDF

Copyright (c) 2026 International Journal of Intelligent Systems and Data Science

This work is licensed under a Creative Commons Attribution 4.0 International License.