Abstractive video lecture summarization: applications and future prospects

Modern educational technology systems allow learners to access large amounts of learning materials such as educational videos, learning notes, and teaching books. Automated summarization techniques simplify the access and exploration of complex data collections by producing synthetic versions of the original content. This paper addresses the problem of video lecture summarization by means of abstractive techniques. To enhance the accessibility of the video lecture content in challenging contexts or while coping with learners with special needs it produces a synthetic textual summary condensing the key concepts mentioned in the lecture's speech. Unlike prior works based on extractive methods, the proposed method can produce more readable and actionable summaries, not necessarily composed of existing portions of speech content. To compensate the lack of annotated data, it also opportunistically reuses the pretrained models available for meeting summarization. The experimental results achieved on a benchmark dataset show that the proposed method generates more fluent and actionable summaries than prior approaches simply relying on content extraction. Finally, we also envision further applications of summarization techniques to learning content. The future prospects of use of summarization techniques in education have shown to go well beyond video summarization.