Extracting Metadata from Learning Videos for Ontology-Based Recommender Systems Using Whisper & GPT

In modern education, individualized learning environments play a vital role by allowing learners to tailor their learning paths based on personal needs, interests, and abilities. Achieving effective individualization relies on dynamic adaptation of the learning path, typically facilitated by recommender systems. These systems offer personalized suggestions, commonly employing content-based or collaborative filtering approaches. However, traditional recommender systems often lack consideration of the semantics of learning elements. To address this limitation, ontology-based recommender systems integrate semantic modeling, establishing additional connections within a domain to enhance precision and context in recommendations. Notably, these systems mitigate the cold start problem and are particularly advantageous in learning environments with limited data. While videos are prevalent in learning platforms, their unstructured nature poses challenges for processing. This paper introduces an innovative approach, leveraging Large Language Models, specifically GPT, to extract metadata from learning videos. The proposed method intelligently augments videos and links them to a domain ontology, enabling the integration of videos into ontology-based recommender systems. The application of this approach is demonstrated through a case study in software engineering education, showcasing its potential to enhance individualized learning experiences in specific domains. The presented method offers an automated alternative to manual video processing, aligning with the evolving landscape of education technology.