Machine learning-based collaborative learning optimizer toward intelligent CSCL system

Recently, various kinds of collaborative learning have been attempted. However, since there are many collaboration patterns, it is difficult for teachers to identify good collaborations among the learners. For carrying out good collaborative learning, it is desirable that teacher finds out the good collaborative patterns among the learners. To develop a CSCL system for solving these problems, a questionnaire survey was performed for the possibility of predicting understanding level from the learners' collaboration. We measured the learners' personalities, the number of collaborated people and the understanding levels. By using machine learning with the obtained data, we attempted to develop a prediction model for understanding level. We measured a generalization scores of it by using test data. The generalization scores of the prediction model were 0.60 ∼ 0.70. Moreover we proposed a method to estimate the optimal number of collaborating people, named “Collaborative Learning Optimizer (CLO)”. We showed a possibility for the prediction of the optimal number of the collaborating people from learner's personality.