The AERA dataset comprises noisy assessment rationales generated from large language models (LLMs), designed to enable explainable student answer scoring. It specifically targets science and biology questions from the publicly available The Hewlett Foundation: Short Answer Scoring competition. Further data creation and training details can be… See the full description on the dataset page: https://huggingface.co/datasets/jiazhengli/AERA.