Knowledge graph convolutional networks with user preferences for course recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Educational Technology

Background:

• Massive Open Online Courses (MOOCs) are rapidly expanding, creating challenges in personalized course discovery.
• Knowledge Graphs (KGs) offer rich semantic information to improve recommendation systems, especially addressing data sparsity in MOOCs.
• Existing KG-enhanced recommendation algorithms highlight the importance of utilizing side information for accuracy.

Purpose of the Study:

• To introduce KGCN-UP (Knowledge Graph Convolutional Networks with User Preferences), a novel model for predicting user-course interaction likelihood.
• To enhance user representation and item representation by leveraging relational chains and semantic relationships within KGs.
• To address data sparsity and improve the quality of MOOC recommendations.

Main Methods:

• Developed KGCN-UP with two key modules: user preference propagation and item neighbor enhancement.
• The user preference module refines user preferences via relational chains and dynamic attention in the KG.
• The item neighbor module enhances item representations by aggregating semantic relationships and relationship-type-based attention weights.

Main Results:

• KGCN-UP effectively leverages high-order structural and semantic information from KGs.
• The model significantly outperforms existing state-of-the-art recommendation models on a real-world dataset.
• Empirical results demonstrate substantial improvements in recommendation accuracy.

Conclusions:

• KGCN-UP offers a powerful approach to personalized course recommendation in the MOOC landscape.
• The model's ability to refine user preferences and enhance item representations addresses key challenges in recommender systems.
• The findings suggest that KGCN-UP can significantly improve user satisfaction and engagement with online learning platforms.