Curriculum-Guided Reinforcement Learning for Adaptive Educational Game Platforms

Abstract

Adaptive educational technologies offer personalized learning pathways, but traditional recommendation schemes often fail to balance engagement with long-term concept mastery. Reinforcement learning (RL) offers a promising direction by optimizing policies that consider delayed rewards, yet RL approaches in education face significant sample inefficiency and reward sparsity. This research integrates curriculum sequencing principles into RL-based educational game platforms. The system constructs a knowledge graph of interconnected learning objectives and uses it to shape the reward function for a policy that adapts the sequence of activities presented to the learner. A study involving 2,150 students over a semester compared this curriculum-guided RL against static sequencing and collaborative filtering baselines. Learners guided by the proposed approach demonstrated a statistically significant increase in retention rates (17.4%) and task completion efficiency (22.5% reduction in time-on-task) over the control groups. The results suggest that combining domain knowledge with RL can improve adaptive learning systems’ effectiveness, offering a path toward scalable personalization.