Retrieval-Augmented Generation for Finding Relevant Lectures from Quizzes in a Multilingual STEM Educational Environment

Understanding the semantic connections between the increasingly online-based offered educational content is becoming more relevant for educational institutions. The fast adoption of AI-based techniques and their application in semantic search offers new possibilities and challenges when working in an educational context. In this paper, we explore the usage of broadly used text embedding models in Retrieval-Augmented Generation (RAG) systems for semantic retrieval when searching multilingual educational content from Massive Open Online Courses (MOOCs), a significant part of which belongs to engineering studies. More specifically, we analyse the performance of text embedding models when searching multilingual content and the impact of mathematical notation on the retrieval results. Furthermore, we introduce a series of developed prototypes and applications that inspect embedding models and utilise them to find semantically relevant lectures and courses.