Automated exam script checking using zero-shot LLM and adaptive generative AI

Abstract

The manual grading of exam scripts is a labor-intensive process, often plagued by subjectivity and inconsistency. This study presents an innovative approach to automating the evaluation of exam scripts using the Large Language Model (LLM) integrated with Zero-Shot Learning (ZSL) and Generative AI. Our system leverages the capabilities of GPT-4 to generate and evaluate answers, applying similarity measures to ensure accuracy and fairness in grading. The model's adaptability through ZSL allows it to assess new questions without additional training. The system also incorporates iterative refinement techniques to enhance the quality of standard answers over time. Performance evaluations demonstrate a low error rate, with an average relative error of 1.29% for annotated questions and 1.67% for non-annotated questions compared to human graders. The performance underscores the automated system's transformative potential in revolutionizing educational assessments, offering a consistent, fair, and highly efficient alternative to traditional grading methods.