CogChat: An Academic Alert System of STEM Performance with Cognitive Explanations using ChatGPT

This study introduces CogChat, a novel explainable system that integrates Self-Attentive Knowledge Tracing (SAKT) and Large Language Model (ChatGPT 4o) to enhance STEM academic performance diagnosing. Addressing the issues of interpretability and adaptability in existing models, CogChat offers a comprehensive approach to alerting educators about at-risk students, diagnosing cognitive skills, and generating personalized learning reports. The system comprises three key modules: performance prediction, alert generation, and cognitive diagnosis. Validated through real-world case studies, CogChat effectively supports educational interventions by delivering detailed diagnostic reports, pinpointing specific areas of student difficulty, and offering targeted, evidence-based learning resources. These capabilities demonstrate its significant potential to enhance STEM education by fostering personalized learning and enabling timely, data-driven instructional strategies.