KGAN: Knowledge Grouping Aggregation Network for course recommendation in MOOCs. (January 2023)

• Record Type: Journal Article
• Authors: Zhang, Huanyu; Shen, Xiaoxuan; Yi, Baolin; Wang, Wei; Feng, Yong
• Is Part Of: Expert systems with applications. Volume 211(2023)
• Publication Date: 2023-01
• Subjects: Recommender systems -- Knowledge graph -- MOOCs -- Course recommendation; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.118344
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Massive open online courses (MOOCs) are dedicated to providing learners with large-scale and open-access boutique courses. Recently, the course recommendation algorithm in MOOCs has attracted many researchers' attention. Compared with ordinary recommendation scenarios, course interaction density in MOOCs is more tenuous, since it demands lots of time and gumption of learners. Meanwhile, the connections and semantic information among courses are much more diversiform and plentiful. Therefore, knowledge graph enhanced recommendation algorithms are appropriate for addressing the course recommendation problem in MOOCs scenario. According to relevant research on KG-based algorithms, utilization efficiency of edge information seems to be the key to boosting model effectiveness. In this paper, we propose a high-performance course recommendation model named K nowledge G rouping A ggregation N etwork (KGAN), which uses the course graph (a heterogeneous graph for describing the relations between courses and facts) to estimate learners' potential interests automatically and iteratively. More precisely, KGAN constructs an end-to-end recommendation model, which projects the learner's behavior and course graph into a unified space naturally, which alleviates difficulties in course recommendation such as interaction tenuous, course relevance, and intention diversity. In addition, we proposed intra-group and inter-group attention operator, which packages the propagation set according to the relation-links and obtains the corresponding attention priorities of different entities under different paths for constructing a reasonable and explicit encoding of users. We apply the proposed model on the real-world datasets, and the empirical results demonstrate that KGAN outperforms compelling state-of-the-art baselines. Our implementations are available at https://github.com/StZHY/KGAN . Highlights: The characteristics of student interactions in MOOCs are systematically analyzed. An end-to-end recommendation model KGAN is proposed. KGAN introduces a more reasonable graph feature propagation method. The superiority and effectiveness of the algorithm are verified.