Knowledge-aware reasoning with self-supervised reinforcement learning for explainable recommendation in MOOCs

Explainable recommendation is important but not yet explored in Massive Open Online Courses (MOOCs). Recently, knowledge graph (KG) has achieved great success in explainable recommendations. However, the e-learning scenario has some unique constraints, such as learners’ knowledge structure and course prerequisite requirements, leading the existing KG-based recommendation methods to work poorly in MOOCs. To address these issues, we propose a novel explainable recommendation model, namely K nowledge-aware R easoning with self-supervised R einforcement L earning (KRRL). Specifically, to enhance the semantic representation and relation in the KG, a multi-level representation learning method enriches the perceptual information of semantic interactions. Afterward, a self-supervised reinforcement learning method effectively guides the path reasoning over the KG, to match the unique constraints in the e-learning scenario. We evaluate the KRRL model on two real-world MOOCs datasets. The experimental results show that KRRL evidently outperforms state-of-the-art baselines in terms of the recommendation accuracy and explainability.