Multimodal Deep Learning for Classifying Student-generated Questions in Computer-supported Collaborative Learning

The quality of student-generated questions in computer-supported collaborative learning (CSCL) plays a pivotal role in determining the overall effectiveness of CSCL experiences. While prior research has explored the impact of student-generated questions in CSCL, instructors currently face the formidable task of continuously and simultaneously monitoring multiple groups to assess the quality of these questions. This monitoring process places a significant burden on instructors already grappling with the challenges of facilitating CSCL. In response to this challenge, we introduce a novel dataset featuring annotated student-generated questions derived from real CSCL environments, categorized according to Bloom's taxonomy. Utilizing this dataset, we propose M-BloomNet, a multimodal deep learning model that exhibits state-of-the-art classification accuracy in this dataset. Moreover, we demonstrate the practical application of our dataset and model within a real CSCL framework, illustrating their potential to ease the monitoring burden on instructors and enrich the CSCL experience.