EduEVAL-DB
收藏资源简介:
EduEVAL-DB是由马德里自治大学团队构建的面向K-12教育的教学解释风险评估数据集,包含854条基于ScienceQA基准的子集生成的解释文本。数据集涵盖科学、语言艺术和社会科学三大领域,每条问题配备1个人类教师和6个LLM模拟教师角色的解释,并通过半自动专家标注流程标记五类教学风险维度。其创新性在于通过提示工程构建差异化教师角色(如‘热情跑题型教师’‘自信错误型教师’),模拟真实教学场景中的风格与缺陷。该数据集旨在支持AI教学助手和自动教学评估模型的开发,特别关注消费级硬件可部署的轻量化模型在风险检测方面的微调验证。
EduEVAL-DB is a teaching explanation risk assessment dataset for K-12 education, constructed by the team from the Autonomous University of Madrid. It contains 854 explanatory texts generated from the subset of the ScienceQA benchmark. The dataset covers three major domains: science, language arts, and social sciences. Each question is paired with explanations from one human teacher and six LLM-simulated teacher roles, and five types of teaching risk dimensions are annotated via a semi-automatic expert annotation workflow. Its innovation lies in constructing differentiated teacher roles (e.g., "enthusiastic question-racing teacher", "confident but error-prone teacher") through prompt engineering, simulating the teaching styles and flaws in real educational scenarios. This dataset aims to support the development of AI teaching assistants and automatic teaching assessment models, with a particular focus on fine-tuning and validation of lightweight models deployable on consumer-grade hardware for risk detection.



