Reimagining Personalization through Explainable Decision Support for Learners

Abstract

The expansion of digital education has enabled flexibility but also introduced cognitive overload as learners navigate vast ecosystems of resources. Intelligent decision support sys-tems have emerged to mediate this complexity, yet most rely on behavioral correlations rather than pedagogical reasoning, lacking transparency and fostering dependence on opaque mod-els. This article reports findings from a large-scale audit of multiple recommendation ap-proaches applied to diverse educational datasets, analyzing their capacity to guide learners effectively and transparently. Results show that collaborative and autoencoder-based methods generalize well, while knowledge-aware models outperform when curricular structures are explicit. Path-reasoning methods, though not top performers in ranking accuracy, deliver valuable pedagogical explanations that promote reflection, positioning intelligent systems as transparent and pedagogically grounded mediators.