Abstractive Text Summary with Transformer on Youtube Video Subtitle

Abstract

Time limitation is one of the most important factors when consuming media. Longer duration makes it harder for users to watch the entirety of the video. Text summarization could be a way for users to acquire information swiftly and concisely. However, the extent to which the summary of the information made has really approached the main core of the information to be conveyed. In this study using YouTube video subtitles as the data that will be used to get a summary of the core information from the video. Consequently, this research focuses on abstractive summarization utilizing several Transformer models namely T5, BART, and PEGASUS, and using the video subtitle dataset to create a summary. The text data from the video subtitle is used as the main source of information in the learning process of the model, ultimately enhancing the model’s ability on this specific summarization task. In evaluating the models’ results, ROUGE is employed, specifically ROUGE-1, ROUGE-2, and ROUGE-L.