Reinforcement Learning for Affect Aware Content Adaptation in Intelligent Tutoring Systems

Intelligent Tutoring Systems (ITS) have advanced in personalized learning, but their ability to adapt to learners' emotional states is limited. This paper proposes an emotion-aware content adaptation framework using Reinforcement Learning (RL). A lightweight CNN extracts facial expressions and emotional features from video streams, which are then integrated with cognitive data. A Deep Q-Network (DQN) dynamically adjusts content difficulty, presentation, and pacing to enhance long-term learning outcomes. This approach enables personalized learning experiences based on students' emotional states (such as frustration or boredom), making intelligent tutoring systems more interactive and adaptive, thereby enhancing students' emotional satisfaction and long-term learning outcomes. This framework offers a novel solution for emotion-driven adaptive teaching in intelligent tutoring systems.