<![CDATA[Manual essay assessment is often time-consuming, subjective, and difficult to implement consistently in large-scale educational settings. Therefore, this study aims to develop and evaluate an integrated AES framework that combines essay scoring, spelling error detection, semantic relevance analysis, and automated feedback generation. The proposed framework integrates BERT for essay score prediction, pyspellchecker for spelling error detection, SBERT for semantic relevance analysis, and T5 for automated feedback generation. Experiments were conducted using a publicly available dataset consisting of 1,785 essays for model development and 592 essays for testing. The BERT model achieved strong predictive performance with an MAE of 0.2381, RMSE of 0.2847, and R² of 0.9513 on the test dataset, indicating high agreement with human scoring. The spelling detection module successfully identified linguistic errors, while the SBERT model produced relevance scores ranging from 0.58 to 0.95. The T5 model generated feedback that was consistent with essay scores, spelling quality, and semantic relevance. These findings demonstrate that integrating multiple NLP components can extend AES functionality beyond score prediction by providing comprehensive writing evaluation and formative feedback. Practically, the proposed framework supports scalable essay assessment and automated feedback generation. However, its generalizability is limited on a single-topic dataset and synthetic feedback data.]]>
Copyrights © 2026
