Key points are not available for this paper at this time.
Abstract This paper proposes an automated system for grading handwritten subjective answers, leveraging advanced computer vision, natural language processing, and large language model techniques. Although time-consuming, the system presents a promising theoretical approach by employing CRAFT for text detection, TrOCR for handwritten text recognition, and a fine-tuned language model for answer evaluation. Experimental results demonstrate the system's potential accuracy in transcribing handwritten text and consistency in grading answers compared to human raters. The proposed methodology offers a scalable and efficient solution to automate the traditionally labor-intensive task of grading handwritten responses, with the potential to transform education assessment practices. The system's performance, limitations, and future research directions to improve efficiency are discussed.
Hiremath et al. (Thu,) studied this question.