A method called Collaboration Scenario-based Scale for Emotion Regulation (CSSER) was proposed for evaluating coordination ability in collaborative learning. The method presents cartoons of several scenarios that occur in collaborative learning to the learners, and then they are asked to fill in blank balloons with their thoughts and statements about some of the scenarios; their responses are evaluated manually based on a rubric. In this paper, we propose a BERT-based classification model for automatic text scoring in CSSER, which enables us to group learners based on their abilities measured by CSSER. Further, we demonstrate the effectiveness of data-augmentation techniques for the proposed method using lexical substitution and back-translation.