<![CDATA[Automated Essay Scoring (AES) using Large Language Models (LLMs) is rapidly evolving, yet there remains a lack of studies comparing Zero-shot prompting and Fine-tuning approaches on open-weight LLM models. This study analyzes the comparison of these two approaches on the Qwen 2.5-1.5B-Instruct model using the ASAP-AES 2.0 dataset. The Zero-shot approach employs Role Prompting and Chain-of-Thought (CoT) techniques, while Fine-tuning uses Parameter-Efficient Fine-tuning (PEFT/LoRA). Evaluation was conducted using the Quadratic Weighted Kappa (QWK) and Root Mean Square Error (RMSE) metrics. The results show that the Zero-shot approach yields a QWK of 0.0069 with an extreme central tendency bias, while Fine-tuning yields a QWK of 0.3247 (fair agreement). These findings confirm that adapting a Fine-tuning approach is one way to produce a sufficiently accurate automatic essay grading system on small-scale open-weight models]]>
Copyrights © 2026
