A multiplex graph neural network framework for educational collaboration and influence analytics: a case study

Social networking has become a promising area of research and a medium to boost innovation and learning at departmental and educational levels. The behavior of individuals in social networking sites often follow complex social dynamics which cannot be captured using traditional methods and approaches. This study presents a comprehensive case study of the Department of Artificial Intelligence and Data Science at Parul University, India, to understand collaboration dynamics and influence structure. A temporal multiplex network is constructed by collecting data from different sources and timeframes spanning a period of five years. To study the collaboration dynamics within the department, a combination of network measures and machine learning approaches such as Graph Neural Networks and Temporal Graph Attention Models are employed. Three research questions are analysed in this study: (RQ1) What structural properties characterise the collaboration network? (RQ2) How do collaboration and influence patterns evolve over time? (RQ3) Which network signatures predict future influence emergence? These questions are operationalised through three analytical tasks: (Task 1) temporal link prediction for collaboration formation, (Task 2) classification and regression for influential actor identification, and (Task 3) dynamic community detection for role transition analysis. The generated network was found to be a small world with an increase in structural cohesion. Five distinct influence archetypes were identified with strong behavioral profiles. The model performed well in predicting link formation and influence emergence. Diversification and brokerage in early years predict the likelihood of individuals following one of four different long-term research career trajectories. In methodological advance, we present a reproducible method for temporal multiplex network analysis that can be applied in a variety of educational contexts. The findings have implications for career management of academics, identifying future leaders, targeting resources to positions that yield the greatest value by filling structural holes, and designing pedagogy for productive collaboration. However, it is important to emphasise that our predictive models establish statistical associations, not causal relationships. Collaboration networks in educational settings exhibit predictable patterns that can inform data-driven interventions, though causal validation would require experimental or quasi-experimental designs.