Cross-Scale Collaboration and Thematic Convergence in Facial Recognition Research: Integrating Co-authorship and Keyword Networks, 2010–2025

Abstract

This paper examines how collaboration structure and thematic organisation co-evolve in a large research domain, using facial recognition research as a case study. Drawing on 58,060 Scopus-indexed publications from 2010–2025, we develop an integrated multi-layer analytical pipeline. First, co-authorship networks are constructed at country and institution levels, enabling comparison across levels of aggregation. Second, keyword co-occurrence communities are detected across cumulative longitudinal windows (2010–2015, 2010–2020, and 2010–2025) to identify changes in the thematic structure of the field. Third, in a projection step, these keyword communities are used to classify the publication records of the most central network actors, allowing thematic distributions to be compared across countries and institutions. Three main findings emerge. Country-level integration coexists with substantially greater institutional fragmentation, revealing a marked cross-scale structural asymmetry. Longitudinally, established biometric and feature-engineering communities persist while deep-learning approaches become increasingly central. Finally, the most central countries and institutions display highly similar thematic profiles concentrated around three dominant communities. The study demonstrates how collaboration networks and keyword networks can be integrated to analyse cross-scale structure, longitudinal change, and thematic convergence within large scientific fields.