Learning how to teach from “Videolectures”: automatic prediction of lecture ratings based on teacher's nonverbal behavior

摘要

Large repositories of presentation recordings (e.g., “Videolectures” and “Academic Earth”) often provide their users with rating facilities. The rating of a presentation certainly depends on the content, but the way the content is delivered is likely to play a role as well. This paper focuses on the latter aspect and shows that nonverbal behavior (in particular arms movement and prosody) allows one to predict whether a presentation is rated as low or high in terms of quality. The experiments have been performed over 100 presentations collected from “Videolectures” and the accuracy is up to 66% depending on the techniques adopted. In other words, nonverbal communication actually influences the ratings assigned to a presentation.