EasyDKT: an easy-to-use framework for Deep Knowledge Tracing

The goal of knowledge tracing (KT) is to track a students’ progress over time by analyzing their historical data, so as to predict their future performance on tests related to the topics they have covered. The rise of online platforms for education, where the learning process is embedded, unlocked the potential of customized teaching such as in intelligent tutoring systems. Thanks to ongoing advancements in KT algorithms, teachers can now be aware of students’ needs and recommend appropriate learning resources. They can also rank learning content, skipping or delaying content based on difficulty. In recent years, Deep Knowledge Tracing (DKT) has proven highly effective in solving KT tasks due to its ability to model complex long-range dependencies in test sequences, resulting in better prediction quality. The field of DKT is expanding, with numerous algorithms being proposed and implemented using various technologies. This paper introduces a new framework called EasyDKT, which simplifies the development and evaluation process for DKT algorithms. The framework aims at offering users a high level of technological abstraction, with a modular structure that considers data processing, evaluation metrics, and neural network models to be trained on custom datasets. Currently, EasyDKT supports PyTorch and TensorFlow, with plans to incorporate additional technologies in the future. Experiments on the ASSISTments skill-builder dataset 2009-2010 show a case study of students’ data analysis through EasyDKT.