Core Concepts
Knowledge tracing aims to monitor students' evolving knowledge states during the learning process and predict their performance on future exercises, enabling the provision of more intelligent educational services.
Abstract
This survey provides a comprehensive overview of knowledge tracing (KT), a fundamental task for analyzing student behavioral data in online education. It first introduces three types of fundamental KT models: Bayesian models, logistic models, and deep learning models. Bayesian models employ probability models, logistic models use logistic functions, and deep learning models leverage neural networks.
The survey then reviews extensive variants of these fundamental KT models that consider more stringent learning assumptions, such as modeling individualization before learning, incorporating engagement during learning, considering forgetting after learning, and utilizing side information across learning. These variants aim to reflect a more comprehensive learning process in real-world scenarios.
Additionally, the survey presents typical applications of KT in various educational scenarios, including learning resources recommendation, adaptive learning, and broader applications beyond student learning. To facilitate research and development in this field, the authors have also developed two open-source algorithm libraries: EduData for downloading and preprocessing KT-related datasets, and EduKTM for providing extensible and unified implementations of existing mainstream KT models.
Finally, the survey discusses potential future research directions in this rapidly growing field, with the goal of assisting both researchers and practitioners in fostering the development of KT and benefiting a broader range of students.
Stats
The survey does not contain any specific statistics or figures to extract.
Quotes
The survey does not contain any striking quotes to extract.