Enhancing Automated Essay Scoring with Three Techniques: Two-Stage Fine-Tuning, Score Alignment, and Self-Training
Hongseok Choi, Serynn Kim, Wencke Liermann, Jin Seong, Jin-Xia Huang
TL;DR
This work tackles data scarcity in Automated Essay Scoring by proposing a modular trio of techniques—Two-Stage fine-tuning with LoRA, Score Alignment, and Uncertainty-aware Self-Training—implemented on the DualBERT architecture. The methods are designed to be complementary, boosting performance in limited-data scenarios while preserving or enhancing full-data results; empirically, Score Alignment alone delivers state-of-the-art performance in the full-data setting, and the integrated pipeline achieves 91.2% of full-data performance in a 32-data regime on ASAP++ by leveraging unlabeled data with uncertainty-aware filtering. The approach demonstrates strong data-efficiency, stability across prompts, and practical potential for cross-prompt AES with limited labeled resources. Overall, the paper advances AES by providing a flexible, modular framework that improves robustness and scalability in real-world, data-constrained educational environments.
Abstract
Automated Essay Scoring (AES) plays a crucial role in education by providing scalable and efficient assessment tools. However, in real-world settings, the extreme scarcity of labeled data severely limits the development and practical adoption of robust AES systems. This study proposes a novel approach to enhance AES performance in both limited-data and full-data settings by introducing three key techniques. First, we introduce a Two-Stage fine-tuning strategy that leverages low-rank adaptations to better adapt an AES model to target prompt essays. Second, we introduce a Score Alignment technique to improve consistency between predicted and true score distributions. Third, we employ uncertainty-aware self-training using unlabeled data, effectively expanding the training set with pseudo-labeled samples while mitigating label noise propagation. We implement above three key techniques on DualBERT. We conduct extensive experiments on the ASAP++ dataset. As a result, in the 32-data setting, all three key techniques improve performance, and their integration achieves 91.2% of the full-data performance trained on approximately 1,000 labeled samples. In addition, the proposed Score Alignment technique consistently improves performance in both limited-data and full-data settings: e.g., it achieves state-of-the-art results in the full-data setting when integrated into DualBERT.