GCCR: GAT-Based Category-Aware Course Recommendation

Massive Open Online Courses (MOOCs) are widely receiving attention from learners because of their ability to recommend personalized courses to learners. While existing course recommendation methods have shown good performance, they still overlook the learning sequence from easy to difficult and the correlation between different disciplinary categories, resulting in unsatisfactory performance. To tackle these, we propose a GAT-Based Category-aware Course Recommendation system, named GCCR. Specifically, we integrate the learner-course-category tripartite graph and course sequence graph into a unified large-scale graph, and then introduce a learner-centered graph sampling strategy within this unified graph, optimizing GAT-based propagation. Besides, we utilize the genetic algorithm to adaptively search for a locally optimal assignment plan based on the correlation between disciplinary categories. Finally, to improve model convergence stability, we introduce the weighted dual prediction, assigning distinct weights to the losses of GCCR and its GAT-less variant to train simultaneously, named GCCR + D. Extensive experiments on real-world datasets from two different periods demonstrate the effectiveness of our model compared to prevailing baseline methods.