Explainable Multimodal Student Profiling and Personalized Course Recommendation using Attention-Enhanced Heterogeneous Graph Neural Networks

The rapid expansion of digital learning environments has generated rich and diverse student data, yet many existing academic support systems still rely on unimodal predictors that overlook the relational nature of learning. This work introduces an Attention-Enhanced Heterogeneous Graph Neural Network (HGNN) that unifies multimodal student profiling and personalized course recommendation within a single explainable framework. The educational ecosystem is modelled as a heterogeneous graph composed of students, courses, and socio-academic attributes, where distinct relational edges capture course enrolment patterns, peer similarity, and contextual demographic influences. An edge-type-aware attention mechanism enables the model to selectively emphasize the most influential relationships, thereby offering transparent justification for each prediction. Using a real institutional dataset of 400 learners across multiple semesters, the proposed framework achieved 94% classification accuracy surpassing conventional machine learning baselines and homogeneous graph models. The analysis of the attention revealed valuable academic and behavioral variables, which create the learning trajectory of each student. This research takes the field a step closer to trustworthy, context-sensitive, and practical educational analytics by combining classification, recommendation, and interpretability into one pipeline. It prepares the ground for early intervention, improved decision-making, and academic guidance on an individual scale.