Enhancing Personalized Recommendation via Metacognitive Profile

Abstract

Personalized recommendation systems are increasingly applied to massive open online course platforms to alleviate information overload. Recent advancements have leveraged graph neural networks to model user-course relationships and further enhance the accuracy of recommendations. However, most methods overlook the high-order metacognitive skills, which can influence users’ choices, resulting in biased predictions. Therefore, we introduce MetaGAN, which first integrates users’ metacognitive skills into course recommendations. Specifically, we extract metacognitive skills from user learning behaviors and construct the user metacognitive profile. Furthermore, we introduce an attention-based feature fusion module to integrate the metacognitive profile and course embeddings effectively. Finally, we propose a user metacognitive profile loss to further increase the recommendation performance. Comprehensive experiments on two real-world datasets show that MetaGAN achieves state-of-the-art performance in online course recommendation tasks. Our code is available at https://github.com/yinjiaqi/MetaGAN.git.