KT4EQG: Personalized Exercise Question Generation via Knowledge Tracing

KT4EQG: Personalized Exercise Question Generation via Knowledge Tracing bridges two key AI-in-education paradigms: Personalized Learning through question generation and Learning Analytics through knowledge tracing. Rather than generating generic practice questions, KT4EQG uses a Knowledge Tracing model to first identify the knowledge concept that would maximize a student's potential improvement in overall mastery, then trains an LLM-based generator to produce a question faithfully grounded in that concept. This two-stage architecture — KT for concept selection, LLM for faithful question generation — outperforms less personalized baselines on XES3G5M and MOOCRadar datasets. The approach represents a significant advance in Adaptive Learning system design, connecting to Knowledge Tracing IRT research on modeling student knowledge states and Automated Question Generation work on producing high-quality educational content. Unlike earlier systems such as Slidesqaqa Pedagogical Question Generation that generate questions from static content, KT4EQG personalizes based on dynamic student models, aligning with Intelligent Tutoring goals of providing the right question at the right time for each learner.

Connected Concepts

Connected Articles

Citation

Xinyi Gao, Qiucheng Wu, Lu Ding, Q. Vera Liao, Kaizhi Qian, Ying Xu, Shiyu Chang, Yang Zhang (2026). KT4EQG: Personalized Exercise Question Generation via Knowledge Tracing. arXiv:2605.23933. arXiv preprint.