Researchers have developed a novel approach to enhance Automated Essay Scoring (AES) systems, particularly in scenarios with limited labeled data. The proposed method integrates three techniques: a two-stage fine-tuning strategy using low-rank adaptations, a score alignment method for improved consistency, and uncertainty-aware self-training with unlabeled data. Experiments on the ASAP++ dataset demonstrated that these techniques, when combined, achieved 91.2% of the performance of a full-data trained model, even with only approximately 1,000 labeled samples. The score alignment technique, in particular, showed consistent performance improvements and achieved state-of-the-art results in the full-data setting when applied to the DualBERT model. AI
IMPACT These techniques could significantly improve the scalability and accessibility of AI-powered educational assessment tools, especially in resource-constrained environments.
RANK_REASON The cluster contains an academic paper detailing novel techniques for a specific AI application (Automated Essay Scoring). [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
