Selection of key sentences from lecture video transcription and its application to feedback to the learner
Miki Takeuchi, Akinori Ito, Takashi Nose · 2024
The COVID-19 pandemic accelerated the shift to online university lectures. Real-time online lectures offer live interaction, while on-demand video lectures such as MOOCs allow flexibility; however, on-demand video lectures face low student concentration and engagement challenges. Thus, our project aims to develop an interactive agent for on-demand videos, improving student motivation. As a building block of such a system, we develop a method to estimate key sentences from lecture speech. We compared two text summarization methods, the BERTSUM and GPT-based summarization, to select the key sentences. As a result, a combination of GPT-based summarization and BERT-based text selection gave the best results. Moreover, we developed a system that indicates the estimated key sentences to the learners and conducted a subjective evaluation of the system.