Answers are grounded in this paper and cite their sources.
Automated essay scoring demonstrates improved accuracy and discrimination in scoring levels using a novel AI model.
Automated essay scoring is of great practical value in educational applications, yet existing methods heavily rely on labeled data and insufficiently distinguish semantic differences across score levels. To address these issues, we propose a BERT (Bidirectional Encoder Representation from Transformers) based Deep Self-Supervised Contrastive Learning Network (D2SCLN). The model employs BERT and a Transformer encoder to learn document-level representations, optimized via supervised classification. To better exploit unlabeled data, a self-supervised reconstruction mechanism is introduced to enhance representation robustness. In addition, supervised contrastive learning is incorporated to improve discrimination among different scoring levels. Experiments on the ASAP dataset show that D2SCLN consistently outperforms representative baselines under multiple data splits in terms of Accuracy, AUC, and AUPR, demonstrating its effectiveness and stability.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: