Automated Essay Grading of Constructive Response Test Responses for Mechanical Engineering Students

Machine learning education related applications have increased with the appearance of large language models. While automatic essay grading (AEG) has been studied extensively in the past, most of these studies have focused on evaluating English competence instead of assessing knowledge competence in an engineering field. This study aimed to develop an AEG model to evaluate student’s mechanical engineering Constructive Response Test (CRT) question responses which were instructor graded. Because of the small number of student responses (45), a synthesized set of responses was also generated by using text-to-text paraphrasing models. A neural network grading engine was built and trained to assess comprehension utilizing the Bidirectional Encoder Representation Transformer (BERT) and related models on student and synthesized responses. This study showed that the AEG based Natural Language Processing (NLP) model showed high accuracy and a higher degree of consistency in grading student responses compared to instructor-graded responses.