ASAGeR: Automated Short Answer Grading Regressor via Sentence Simplification

Mohammad Iqbal, Rosita Laili Udhiah, Tsamarah Rana Nugraha, Hsing-Kuo Pao

Abstract: We propose an automated short answer grading system (ASAG) to estimate the student answer scores via text summarization from LLMs. The step of text summarization provides enough question answer normalization so that the summarized answers have the answer keys well organized and the grading based on that should be more accurate and easier than before, no matter the answers are graded by human or automatic graders. On the other hand, we also discuss the scenario when more than one grader are involved in the grading but providing inconsistent scores. We adopt a majority voting mechanism to overcome such difficulty and produce superior result in average. Overall the proposed methodology has its evaluation done to show the superiority to other state-of-the-art methods. The pre-trained transformer version 3.5 (GPT 3.5) is used to serve the text summarization tool given a well-designed prompt.