A student modeling method combining Deep Learning and forgetting models with Knowledge Tracing
Shinichi Oeda, Shunichi Hasegawa · Procedia Computer Science · 2023
Educational Data Mining (EDM) aims to enhance education by analyzing learners’ skills and question difficulty levels using machine learning methods. Knowledge Tracing (KT), a subfield of EDM, utilizes Hidden Markov Models to estimate learners’ abilities and predict their performance on unseen questions. While deep learning methods, such as bi-directional RNNs, have improved KT's accuracy, they may lack interpretability from an educational psychology perspective. Item Response Theory (IRT), widely used in educational statistics, offers greater explanatory potential. This study proposes a model that integrates the concept of forgetting into IRT for improved accuracy and explainability in Knowledge Tracing using bi-directional RNNs. The forgetting concept is based on Ebbinghaus’ forgetting curve theory. Three experiments were conducted using synthetic data to compare a model from a previous study, a model based on the proposed method, and a model that combines the previous study's model with IRT but excludes the forgetting concept.