Automated-Computer Based Assessment of Free-Text Exam Answers by Transformers

In this study, we aim to develop a Python module that leverages transformer technologies to automatically grade text-based open-ended questions. The proposed approach utilizes natural language processing techniques to compare student responses with the solution key. Initially, both the solution key and student answers are converted into vector representations using pre-trained models. Then, a score is assigned to each student response based on the similarity and overlap between the vectors. To evaluate the effectiveness of this approach, a selection of questions is graded by both the module and human instructors, with the results being compared to assess the accuracy and reliability of the automated grading system.