Samenvatting

The recent advancements in machine learning (ML) techniques have significantly improved the performance in various natural language processing (NLP) tasks, particularly in deep learning with the transformer architecture. Despite their promising applications in different domains, these innovations have yet to be fully utilized in the education sector. This can be attributed to the large amount of data required for these ML techniques, which schools may not have access to. This thesis aims to address this gap by answering the following question: How can semi-structured open-ended questions in student homework assignments automatically be classified by a model that is trained on a small dataset (n=~100) using modern ML techniques? A literature review of commonly used ML techniques is conducted, followed by the development of a prototype for a specific case study. The results demonstrate the viability of automatic classification of homework assignments, although performance is suboptimal for certain types of questions. A RobBERTa-like model is employed for classifying 7 different types of semi-open questions, with promising results observed for 'less complex questions'. The results suggest a correlation between question complexity and classification accuracy. A set of guidelines is proposed to aid future developers in applying similar techniques to their own datasets.