ATHGT-DQ: An Adaptive Temporal Heterogeneous Graph-Transformer Framework for Forecasting Institutional Digital Quotient and Prescribing AI-Enabled Smart University Transformation

Abstract

Institutional digital transformation is increasingly central to the evolution of higher education, yet most existing maturity frameworks remain descriptive, cross-sectional, and weakly adaptive to technological change. This study proposes ATHGT-DQ, an Adaptive Temporal Heterogeneous Graph-Transformer framework for forecasting an institution's Digital Quotient (DQ) and translating forecasts into feasible transformation actions. The framework first operationalizes DQ through six dimensions—digital infrastructure, institutional governance, digital pedagogy, human digital competence, research and innovation ecosystem, ande-administration—and constructs a composite DQ target through a hybrid AHP–entropy–PCA weighting mechanism. To avoid circular prediction, the model forecasts future DQ from lagged institutional indicators rather than reconstructing a same-period weighted score. ATHGT-DQ then combines FT-Transformer embeddings for nonlinear indicator interactions, an HGT layer for typed relationships among institutions, dimensions and indicators, and a Temporal Fusion Transformer for longitudinal dynamics and uncertainty-aware multi-horizon prediction. An ADWIN-based concept-drift module monitors residual distributions and triggers controlled continual updating when the institutional data-generating process changes. SHAP is used for global and local attribution, while constrained counterfactual optimization identifies the smallest feasible indicator changes needed to approach a desired DQ level. Validation is designed around institution-disjoint and time-ordered evaluation, static and deep-learning baselines, component ablations, robustness tests, and statistical significance analysis. The resulting architecture advances digital-maturity research from static measurement toward adaptive forecasting and prescriptive decision support for AI-enabled smart-university transformation.