MOOPer: A Large-Scale Dataset of Practice-Oriented Online Learning

Abstract

With the booming of online education, abundant data are collected to record the learning process, which facilitates the development of related areas. However, the publicly available datasets in this setting are mainly designed for a single specific task, hindering the joint research from different perspectives. Moreover, most of them collect the video-watching or course-enrollment log data, lacking of explicit user feedbacks of knowledge mastery. Therefore, we present MOOPer, a practice-centered dataset, focusing on the problem-solving process in online learning scenarios, with abundant side information organized as knowledge graph. Flexible data parts make it versatile in supporting various tasks, e.g., learning materials recommendation, dropout prediction and so on. Lastly, we take knowledge tracing task as an example to demonstrate the possible use of MOOPer. Since MOOPer supports multiple tasks, we further explore the advantage of combining tasks from different areas, namely, Deep Knowledge Tracing and Knowledge Graph Embedding. Results show that the fusion model improves the performance by over 9.5%, which proves the potential of MOOPer’s versatility. The dataset is now available at https://www.educoder.net/ch/rest.