Towards Automatic Arabic Test Scoring: Leveraging ChatGPT for Prompt Design
Seham Alsulami, Amal Alsubhi, Abrar Alabdli, Tarik Alafif, Mohammad S. Jassas, Khaled Albishre · 2025
ChatGPT possesses powerful large language models that rely on deep learning technology. It excels at understanding natural language and context, enabling it to generate answers to questions and inquiries. However, evaluating and grading high school Arabic Achievement tests in Saudi Arabia requires human interventions by providing correct test answers into the tests platforms. In this paper, we propose a web-based system (AutoGrader), which includes a framework that automatically grades the students' tests without human interventions. The framework relies on the ChatGPT and Term Frequency-inverse Document Frequency (TF-IDF) algorithm. The framework compares students' answers with those generated by the ChatGPT using the TF-IDF similarity measure and then displays students' grades rapidly. The framework is evaluated using students' tests. The framework is efficient and effective, which can be used to improve and enhance educational assessments and experiences.