AIriskEval-edu
收藏资源简介:
AIriskEval-edu数据集由马德里自治大学等机构创建,旨在评估K-12教育场景中教学解释的潜在教学风险。该数据集包含1,639条教学解释,对应170个来自ScienceQA的精选K-12问题,共计8,195个二元标签,其中785个带有可解释性注释。数据来源包括人类教师参考解释和通过Gemini 3.1 Pro API生成的合成解释,这些解释基于六种模拟教师配置文件生成,并经过半自动标注和约30%的教师人工审核。该数据集主要应用于教育人工智能领域,用于训练和评估本地大语言模型评估器,以检测教学解释在事实准确性、深度与完整性、焦点与相关性、学生水平适当性和意识形态偏见等五个维度的风险,从而提升AI生成教育内容的安全性与可靠性。
AIriskEval-edu-db2 is a dataset developed by institutions including the Autonomous University of Madrid, specifically designed to evaluate pedagogical risks of AI-generated explanations in K-12 education scenarios. This dataset contains 1,639 explanatory texts derived from 170 curated ScienceQA questions, covering multiple disciplines including science, language arts, and social sciences. Each entry corresponds to one human teacher-written explanation and 11 AI-generated explanations produced by large language models (LLMs) simulating different pedagogical risk configurations. The dataset was constructed via a semi-automated pipeline: first, simulated explanations were generated based on predefined pedagogical risk dimensions and subjected to binary risk annotation; subsequently, expert teachers verified the annotation results, and added structured interpretability annotations including risk localization and risk description for the identified risk cases. This dataset is primarily applied in the educational technology field, aiming to train and evaluate LLM-based pedagogical auditors to automatically detect multi-dimensional risks in pedagogical explanations, including factual accuracy, depth and completeness, focus and relevance, appropriateness for student levels, and ideological bias, thereby enabling monitoring and quality assurance of instructional content in digital learning environments.
AIriskEval-edu 数据集概述
基本信息
- 数据集名称:AIriskEval-edu
- 来源地址:https://github.com/BiometricsAI/AIriskEval-edu
- 当前状态:论文正在审稿中(Paper under review)
说明
- 该数据集处于早期公开阶段,论文尚未正式发表。
- README文件中暂无详细的数据集描述、内容示例、使用方式或领域说明。

- 1AIriskEval-edu: New Dataset for Risk Assessment in AI-mediated K-12 Educational Explanations马德里自治大学·生物识别人工智能实验室; 马德里自治大学·GHIA实验室; 拉斯帕尔马斯大学 · 2026年




