Gaze-Driven Adaptive Learning System With ChatGPT-Generated Summaries

Abstract

Enhancing student engagement and comprehension is crucial for effective learning. However, tracking and improving these dynamic states in real-time remains a significant challenge. This study addresses this gap by integrating real-time engagement prediction from gaze data with an adaptive learning system that utilizes ChatGPT-generated summaries to enhance student engagement and learning outcomes. Our experiment with twenty two (N=22) university students demonstrates the effectiveness of gaze data in predicting real-time engagement levels and the impact of adaptive interventions on student engagement, objective and subjective comprehension, and cognitive load. To predict the self-reported engagement and comprehension levels, two deep neural network models, InceptionTime and Transformers were employed. The Transformers model achieved better outcomes, with an average accuracy of 68.15% in predicting engagement levels across a 5-fold StratifiedGroupKFold cross-validation. The results revealed that the experimental group, which received the AI-driven interventions, exhibited significantly better learning outcomes, higher engagement, and better objective comprehension results compared to the control group. Additionally, we observed strong correlations between gaze metrics, engagement levels, and learning outcomes, suggesting that real-time adaptive interventions can dynamically enhance the educational experience. This study advances the field of educational technology by demonstrating the benefits of integrating gaze tracking and AI in learning environments, laying the foundation for dynamic learning interfaces that adapt to individual engagement levels, potentially improving both comprehension and involvement.

Keywords

• Adaptive learning
• E-learning
• ChatGPT API
• eye-tracking
• real-time engagement
• personalized interventions