Automated Student Group Collaboration Assessment and Recommendation System Using Individual Role and Behavioral Cues

Abstract

Early development of specific skills can help students succeed in fields like Science, Technology, Engineering and Mathematics. Different education standards consider “Collaboration” as a required and necessary skill that can help students excel in these fields. Instruction-based methods is the most common approach, adopted by teachers to instill collaborative skills. However, it is difficult for a single teacher to observe multiple student groups and provide constructive feedback to each student. With growing student population and limited teaching staff, this problem seems unlikely to go away. Development of machine-learning-based automated systems for student group collaboration assessment and feedback can help address this problem. Building upon our previous work, in this paper, we propose simple CNN deep-learning models that take in spatio-temporal representations of individual student roles and behavior annotations as input for group collaboration assessment. The trained classification models are further used to develop an automated recommendation system to provide individual-level or group-level feedback. The recommendation system suggests different roles each student in the group could have assumed that would facilitate better overall group collaboration. To the best of our knowledge, we are the first to develop such a feedback system. We also list the different challenges faced when working with the annotation data and describe the approaches we used to address those challenges.