A Survey on Automating Answer-Sheet Evaluation Using AI Techniques

Abstract

The evaluation of answer sheets has traditionally been a time-consuming and subjective process, posing significant challenges in terms of efficiency, scalability, and fairness. With advancements in artificial intelligence (AI) and natural language processing (NLP), automated systems have emerged as promising solutions to these challenges. This study explores two key ap- proaches for implementing automated answer evaluation: BERT- based semantic analysis and large language models (LLMs) powered by prompt engineering. BERT offers deep contextual understanding and precision in grading responses aligned with predefined answer keys, but its reliance on these keys limits its ability to evaluate creative or non-standard answers. In contrast, LLMs such as GPT-4 extend beyond predefined rubrics, utilizing both answer keys and their reasoning capabilities to assess diverse responses accurately. This paper examines the strengths and limitations of these approaches, highlighting their potential for improving grading accuracy, scalability, and adaptability. By integrating advanced OCR technologies for digitizing handwritten responses, these models can provide a holistic evaluation system. The work em- phasizes the need for flexible frameworks that balance precision and creativity, ensuring fair and efficient evaluation in diverse educational contexts. Through this exploration, we aim to guide the development of scalable AI-driven solutions for modern assessment challenges.