Presentation information

General Session

General Session » [GS] J-2 Machine learning

[1Q3-J-2] Machine learning: structural modeling

Tue. Jun 4, 2019 3:20 PM - 5:00 PM Room Q (6F Meeting room, Bandaijima bldg.)

Chair:Koh Takeuchi Reviewer:Akisato Kimura

4:00 PM - 4:20 PM

[1Q3-J-2-03] Extracting and Exploiting Latent Knowledge Structure by Graph-based Knowledge Tracing

〇Hiromi Nakagawa1, Yusuke Iwasawa1, Yutaka Matsuo1 (1. The University of Tokyo)

Keywords:Deep Learning, Knowledge Tracing, Graph Neural Networks

Recent advancements in computer-assisted learning systems have increased research in the area of knowledge tracing, which estimates student proficiency based on their past performance. In this context, deep learning-based methods, such as Deep Knowledge Tracing (DKT), show remarkable performance; however, existing methods do not consider latent knowledge structure. This limits not only the prediction performance but also the interpretability and validity of models' prediction, which prevents the application to real educational environments. In this paper, we propose a graph-based knowledge tracing model, Graph Knowledge Tracing (GKT). Representing the knowledge structure as a graph, we model students' time-series mastery to each skill using Graph Neural Networks. We consider two problem settings, one is to exploit the pre-defined graph structure and the other is to learn the implicit graph structure from data, and provided two models to deal with them. Using two open datasets, we empirically validated that our method shows higher prediction performance and more interpretable and valid prediction compared to the previous methods. These results show the potential of our proposed method to enhance the performance and the application to real educational environments of knowledge tracing, which could help improve the learning experience of students in more diverse environments.