Uma abordagem automática para extração e classificação de exercícios em vídeos educacionais

Abstract

Videos are widely used in education due to their various benefits, such as facilitating the understanding of complex concepts and promoting student engagement. Recommendation systems can be used to organize and retrieve suitable content based on metadata. Exercises in a video lesson have the potential to provide useful metadata. However, unlike textual sources, there are no clear markers to identify them, as the content is presented in continuous speech by the instructor. Additionally, exercises may be interspersed with tips and explanations provided by the instructor during the speech, which makes identification even more complex. With this premise, this work proposes an automatic classification model for exercises in video lessons based on Digital Bloom Taxonomy, aiming to provide a solid foundation for personalized recommendation systems and expand search possibilities in educational repositories. The evaluation of the proposed model was conducted using classification algorithms, where the BERT model stood out as superior to others.