SIG-Net: GNN based dropout prediction in MOOCs using Student Interaction Graph

Abstract

As the COVID-19 pandemic has increased interest in remote learning, so has interest in massive open online courses (MOOCs). MOOCs have experienced rapid growth, offering the advantages of flexible education without time and space constraints. However, a serious challenge in MOOC learning remains high dropout rates; therefore, significant research efforts have been devoted to predicting student dropouts using deep learning as well as machine learning techniques. In this study, we show that graph neural networks (GNNs) can effectively recognize triadic patterns of interactions among students, objects, and courses, producing effective performance outcomes in MOOC dropout prediction. We propose the Student Interaction Graph Network (SIG-Net) model, which extracts subgraphs from a student interaction graph and predicts dropouts by learning student's interactions in the course that the subgraphs contain. In a student interaction graph, subgraphs provide important information about individual students' interactions in a particular course, which is very helpful for dropout prediction. Our model shows that GNNs can effectively learn from student interaction information in the course, and that student interaction information is very powerful in predicting student dropout. We evaluate the proposed model using two real-world datasets: KD-DCUP 2015 and NAVER Edwith & Boostcourse. This study is the first dropout prediction study using NAVER Edwith & Boostcourse Our evaluation results show that SIG-Net is effective in learning from student interaction, outperforming the well-known machine learning and deep learning models. The source code is available at https://github.com/Noverse0/SIG-Net.git.

Topics & Concepts