In this paper, we propose a method for analyzing the relationship between learners' behavior, which can be observed from learners' video, and their comprehension level. First, we divide a lecture into several spans and in each span we conduct a questionnaire survey about subjective comprehension level of the learners and a quiz on the contents of the lecture. Since the questionnaire survey includes personal differences about their comprehension level and the results of the quizzes are affected by the difference of its difficulty level, these data cannot be used to evaluate the objective comprehension level. We introduce the Item Response Theory and correct these differences. At the same time, we watch the learners' video and sum up the frequency of the expressed behavior for each learner in the video. Then, we classify the learners into cluster based on the frequency of expressed behaviors and analyze the characteristics of the cluster based on the comprehension level in the cluster.
