Non-Linear Navigation in Lecture Videos

Abstract

Massive Open Online Courses (MOOCs) have shown remarkable growth over the past few years. A substantial amount of MOOC content comprises of lecture videos. The major weakness of these lecture videos is the inability to access any content in the video quickly. Participants often use Non-linear Navigation, like skipping, reviewing, multiple- passes, etc. to reach their point of interest in the video. To facilitate quick identification of the point of interest in a video, we propose the design of a system that provides automated lecture video indexing. We introduce an approach to automatically partition the video lecture into segments and present it in the customized video player to the user. The lecture content is organized and presented using features derived from the combination of visual content and audio track of the video to give customized viewing to the learners. To allow non-linear navigation, we generate index points, which indicate the start of a new topic in the video. These index points are created using text extracted from the video by Optical character recognition (OCR) and text from the lecture utterances extracted using Automatic Speech Recognition (ASR). A text-based indexing algorithm is developed to locate these index points. The indexing algorithm merges the neighboring video segments with high text similarity to form a topic segment. Finally, we extract the time-stamp corresponding to the index points and locate it in the video. We evaluated the performance of the system on three hours of video lectures. Experimental results yield 89% indexing accuracy on an average. Further enhancements could improve the accuracy. We believe technologies like this will help efficient navigation of video OER content, especially legacy content. // Paper ID 134