A Knowledge Concept Recommendation Model Based on Contrastive Learning to Improve Knowledge Graph

To meet the fine-grained learning needs of students in massive open online courses (MOOCs), researchers have been increasing attention towards shifting from course recommendation to knowledge concept recommendation. However, existing knowledge concept recommendation methods mainly model the similarity relationships between users or items in heterogeneous information networks (HIN), which suffer from heavy reliance on supervised data, leading to limited representation capability. Based on this, we proposed a recommendation model for knowledge graph improvement based on contrastive learning. First, the model constructs the HIN into a knowledge graph and learns the semantic information of nodes through a propagation-based method. Then, contrastive learning is used to build self-supervised signals to enhance the representation quality of students and knowledge concepts. Finally, the representations of students and knowledge concepts are integrated for recommendation. Experiments on the massive open online dataset MOOCCube showed that the proposed model outperforms the baseline models in all evaluation metrics. Furthermore, the analysis conducted on the computer science and technology dataset of MOOCCube reveals that our model effectively recommends relevant knowledge concepts to students, significantly improving the ranking of these concepts. The outcome demonstrates the substantial practical value of the model.