Cross-domain recommendation in MOOCs: A graph capsule network approach with transfer learning

Cross-domain recommendation systems enhance target domain performance through multi-domain knowledge integration. With the rapid development of artificial intelligence, the growing demand for programming skills has intensified the need for joint recommendations between MOOC platforms and programming resources. However, current approaches predominantly focus on single-domain recommendations, lacking effective mechanisms to address critical challenges including sparse cross-domain user's overlap, dynamic preference modeling, and heterogeneous data integration. To address these limitations, we propose a novel cross-domain recommendation model integrating graph capsule networks with knowledge-aware transfer learning. Our method constructs a heterogeneous network encompassing MOOC and programming domain entities by using cross-domain meta-paths to capture inter-domain semantic relationships. A Hawkes process-enhanced Transformer dynamically is used to encoded temporal shifts in learners' knowledge preferences, while meta-path-guided aggregation extracts global preference features. Visual analysis technology is introduced to assist in the design of meta-paths and provide prior parameters for the network model. Experiments on real-world datasets validate the model's superiority in recommendation accuracy and robustness, demonstrating practical efficacy in authentic educational environments.