jiazhengli/DARS_synthethsis_reflection

DARS反思数据集是一种用于双模型反思评分框架（DARS）训练的合成反思数据集。该数据集包含两个主要文件：reasoner.jsonl和critic.jsonl，分别用于训练生成初始评估并基于反馈进行细化的Reasoner模型，以及提供针对性口头反馈并决定推理何时收敛的Critic模型。数据集采用对话格式，每个条目包括问题提示、学生答案、评估理由以及多轮细化的对话格式。数据集专注于自动化学生答案评分任务，覆盖科学问题和生物考试问题。The DARS Reflection Datasets are synthetic reflection datasets for training the dual-model reflective scoring framework (DARS). The dataset includes two main files: reasoner.jsonl and critic.jsonl, used for training the Reasoner model that generates initial assessments and refines them based on feedback, and the Critic model that provides targeted verbal feedback and determines when reasoning has converged. The dataset uses a conversational format with entries including question prompts, student answers, assessment rationales, and multi-turn conversations for refinement. The datasets focus on Automated Student Answer Scoring (ASAS) tasks, covering science questions and biology exam questions.