MMPDRec: A Denoising Model for Knowledge Concepts Recommendation Using Metapaths

Bohan Zhang, Mo Chen, Jing Chen, Minghe Yu, Zhenghao Liu, Bin Xu, Ge Yu

Abstract: Online education has revolutionized knowledge dissemination, providing unprecedented global access to educational resources. The key to improving online learning environments is effective knowledge concept recommendations tailored to the unique preferences and requirements of each student. While existing GCNs-based recommendation systems contribute significantly to the personalization of content, they frequently neglect the intrinsic relationships between knowledge concepts and struggle to contend with the noise inherent in large-scale educational data, potentially weakening the predictive effect of the model. To address these shortcomings, we propose MMPDRec (Multi-MetaPaths Denoising Recommender System)), an innovative framework that integrates the denoising GCN with a multi-head attention mechanism, considering the diverse reasons students engage with specific knowledge concepts. MMPDRec skillfully captures the subtle patterns in student-concept interactions by utilizing the synergistic impacts of these techniques, yielding a more nuanced understanding that drives the recommendation process. Extensive experiments conducted on a real-world MOOC dataset show that MMPDRec outperforms the state-of-the-art models in predicting and recommending knowledge concepts for intricate online learning scenarios.