Enhancing Knowledge-Concept Recommendations with Heterogeneous Graph-Contrastive Learning

• Published in: Mathematics (Basel) Vol. 12; no. 15; p. 2324
• Main Authors: Wei, Liting, Li, Yun, Wang, Weiwei, Zhu, Yi
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.08.2024
• Subjects: Collaboration; Computational linguistics; Dialectics; Distance learning; Education; Equipment and supplies; graph-contrastive learning; group-structure feature of learner preferences; Homogeneity; Knowledge; knowledge-concept recommendation; Language processing; Medical research; Methods; Natural language interfaces; Online education; Online instruction; Recommender systems; Sampling techniques; Social networks; Sparsity; Students; Teaching
• DOI: 10.3390/math12152324
• ISSN: 2227-7390

With the implementation of conceptual labeling on online learning resources, knowledge-concept recommendations have been introduced to pinpoint concepts that learners may wish to delve into more deeply. As the core subject of learning, learners’ preferences in knowledge concepts should be given greater attention. Research indicates that learners’ preferences for knowledge concepts are influenced by the characteristics of their group structure. There is a high degree of homogeneity within a group, and notable distinctions exist between the internal and external configurations of a group. To strengthen the group-structure characteristics of learners’ behaviors, a multi-task strategy for knowledge-concept recommendations is proposed; this strategy is called Knowledge-Concept Recommendations with Heterogeneous Graph-Contrastive Learning. Specifically, due to the difficulty of accessing authentic social networks, learners and their structural neighbors are considered positive contrastive pairs to construct self-supervision signals on the predefined meta-path from heterogeneous information networks as auxiliary tasks, which capture the higher-order neighbors of learners by presenting different perspectives. Then, the Information Noise-Contrastive Estimation loss is regarded as the main training objective to increase the differentiation of learners from different professional backgrounds. Extensive experiments are constructed on MOOCCube, and we find that our proposed method outperforms the other state-of-the-art concept-recommendation methods, achieving 6.66% with HR@5, 8.85% with NDCG@5, and 8.68% with MRR.