Topic Segmentation of Educational Video Lectures Using Audio and Text

The recent pandemic led to a surge of recorded lecture material available digitally, a resource that can now be used to improve computer-assisted learning. In this paper, we compare two methods for topic segmentation, i.e. the breaking down of a single lecture session into self-contained content units that deal with one or a small set of sub-topics or a set of concepts, respectively. We are interested whether auditory silence or keywords generated by a state-of-the-art keyword extraction tool are superior in segmenting down a session’s recording into self- sufficient clips that may be served to student learners of artificial intelligence. To the best of our knowledge, this is the first comparison of silence-based topic segmentation and keyword-based topic segmentation for recorded lecture materials.