JoWUA

Abstract

Mobile learning environments provide ample opportunity for collecting data from learners, which opens up avenues for creating intelligent and adaptive learning recommendation systems that improve individual learning experiences. Nevertheless, most existing recommendation systems face a challenge in modeling both the semantic relationships between learning concepts and the adaptive pathways that depend on changing learner knowledge states at the same time. In response to the limitations, develop an Adaptive Knowledge Graph Reinforcement Algorithm (AKGRA) for optimizing personalization of learning paths on mobile devices. Method models the structure of an educational knowledge graph along with deep reinforcement learning in an adaptive sequential recommendation setting. The EdNet dataset and learn a semantic knowledge graph that describes the relations of questions, lectures, and educational concepts. Latent semantic dependencies can be learned via TransE-based graph embeddings. The decision for the best recommendation policies is made by a Double Deep Q-Network (DDQN) trained by interactions. The learner state model includes knowledge mastery, engagement behaviors, and graph embeddings in a context-aware manner. Experiment on the EdNet dataset and compare method with conventional recommendation systems, graph-based recommendation systems, and reinforcement learning-based systems. This method, AKGRA, yields 0 0.928, 0.911, 0.937, and 0.921 for precision@10, recall@10, NDCG@10, and MAP, which are significantly superior to other state-of-the-art methods. Statistical analysis shows significant improvements across performance evaluation indices, with p-values indicating strong statistical significance (p < 0.01 to p < 0.001) for most comparisons. Moreover, the convergence behavior of method is stable, the robustness to perturbations is strong, and the scale behavior under larger datasets is excellent.