Automated Essay-Answer Grading using Transformer and GenAI-Based Error Analysis on Answer Sentence

Abstract

Purpose: This study aims to address the limitations of manual essay assessment, which is time-consuming and prone to subjectivity, by developing an automated essay grading system that is faster, more efficient, and more objective. In addition to producing scores, the system is intended to support learning by providing meaningful feedback on students’ writing errors.

Methods: The research applies a transformer-based approach to automated essay scoring in order to capture deeper semantic relationships between student answers and reference answers. Unlike conventional word embedding methods, transformers are used to model contextual meaning at the word, phrase, and sentence levels. The system is further enhanced with an AI-based error analysis module that identifies categories of student errors, including content omissions, conceptual misunderstandings, and structural inaccuracies, enabling the generation of corrective feedback alongside numerical scores.

Result: The experimental results show that the proposed model achieves strong agreement with human scoring, indicated by a Pearson correlation coefficient (r) of 0.8432 and a Mean Absolute Error (MAE) of 0.6013. These results demonstrate an improvement in prediction accuracy compared to word embedding-based approaches. Furthermore, the system successfully generates relevant and actionable error feedback that supports students in understanding their mistakes and improving their essay quality.

Novelty: The novelty of this research lies in the integration of transformer-based automated essay scoring with an AI-driven error analysis module. Rather than focusing solely on score prediction accuracy, the proposed approach combines quantitative scoring with qualitative error feedback, enabling the system to function not only as an assessment tool but also as a learning support mechanism that helps students understand and improve their writing.