1. Introduction
Collaborative learning is a learning style where multiple participants study to acquire knowledge of their learning
subjects1
. The fact that participants obtain a wide variety of educational benefits is supported by many theories, mainly through the collaborative interaction process2
. However, there are also studies that indicate that collaborative
learning does not always bring desirable effects to all participants3 (e.g., participants might arrive at irrational
conclusion if only following the opinion of an egotistic participant). Therefore, technology that can evaluate
collaborative learning situations, such as one that determines whether participants truly study collaboratively, is
important for judging the outcome of collaborative learning.
In collaborative learning performed in face-to-face environments, the participants not only progress in their
learning activity by exchanging utterances, but also by transmitting non-verbal information, such as looking at other
participants. Studies have demonstrated that interaction is important not only for daily conversation, but also for
collaborative learning to maintain smooth communication4, 5. However, to the best of our knowledge, few existing
studies focus on analyzing collaborative learning from the perspective of non-verbal information. Whereas several
studies have been conducted to analyze interaction in terms of non-verbal features in the field of human-computer
interaction6, 7, in order to analyze collaborative learning, such learning aspects as the difference between the
participants’ level of knowledge in their learning subjects should be considered.
In order to clarify the role of non-verbal features in collaborative learning interaction, this study analyzes such
non-verbal interaction among participants. Discussions with others through collaborative means are crucial to
attaining successful learning in collaborative learning. Therefore, we focus in particular on analyzing and estimating
whether the participant attempted to advance the discussion collaboratively (collaborative attitudes). First, to
analyze the non-verbal features that relate to the participants’ collaborative attitudes, we conduct an experiment to
score each of the participant’s collaborative attitudes using the multimodal corpus in collaborative learning collected
by our previous research8
. Then, we verify the effects of non-verbal features and propose an estimation model of the
collaborative attitudes compared to the scores based on multinomial logistic regression analysis.