Researchers have developed a new framework called RLAES that uses reinforcement learning to improve both automated essay scoring and feedback generation. This system incorporates a novel evaluation framework, Rubric-based Feedback Evaluation (RFE), which uses an LLM as a judge to assess feedback quality against fine-grained rubric items. To optimize the training process, Adaptive Gated Feedback Optimization (AGFO) selectively applies these rubric rewards, and Adjacent Contrastive Reasoning (ACR) enhances score calibration by comparing adjacent score levels. Experiments on the ASAP benchmark show RLAES-AGFO achieving state-of-the-art scoring performance among LLM-based methods while maintaining feedback quality comparable to GPT-5.5. AI
IMPACT This research could lead to more sophisticated and nuanced automated feedback systems for educational purposes.
RANK_REASON Academic paper detailing a new method for LLM-based essay scoring and feedback generation. [lever_c_demoted from research: ic=1 ai=1.0]
- Adaptive Gated Feedback Optimization
- Adjacent Contrastive Reasoning
- ASAP benchmark
- GPT-5.5
- RLAES
- Rubric-based Feedback Evaluation
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
