Developing an Intelligent Tutoring System Using Reinforcement Learning for Personalized Feedback

Abstract

The growth of sectors within artificial intelligence have made it feasible to create intelligent tutoring systems (ITSs) that are capable of adapting to specific learner’s needs. This paper reports the implementation of an Intelligent Tutoring System that gives students differentiated feedback utilizing Reinforcement Learning (RL). The primary goal is to improve the learning outcomes by adapting the instruction based on the students’ actions and results. The Q-learning algorithm is employed to maximize relearning on timed feedback control optimizing the timing and content of feedback and assistance provided to the learner and content instruction. The data collection for training and validating the system was done through simulated student environments. The results demonstrate that RL-based tutoring systems are significantly more effective in capturing learners’ attention and improving their test scores compared to nonadaptive, traditional rule-based systems. Furthermore, these results indicate that students experienced less frustration because of the tailored support they received. This study broadens the AI in education knowledge base by demonstrating the efficacy of RL for customizing educational systems. More advanced deep reinforcement learning algorithms will be implemented for future work with goals of shifting domains.