TheMrguiller/ScienceQA
收藏资源简介:
ScienceQA数据集是从小学和高中科学课程中收集的,包含21,208个多模态多选科学问题。其中,10,332个问题(48.7%)包含图像上下文,10,220个问题(48.2%)包含文本上下文,6,532个问题(30.8%)同时包含两者。大多数问题都附有基础讲座(83.9%)和详细解释(90.5%),这些讲座和解释提供了正确答案的一般外部知识和具体原因。ScienceQA是首个大规模多模态数据集,为答案提供了讲座和解释的标注。该数据集涵盖了自然科学、语言科学和社会科学三个领域的丰富多样性,并根据主题、类别和技能进行了详细分类。
The ScienceQA dataset is collected from primary and high school science curricula, consisting of 21,208 multimodal multiple-choice science questions. Of these, 10,332 questions (48.7%) incorporate image contexts, 10,220 questions (48.2%) contain text contexts, and 6,532 questions (30.8%) include both image and text contexts. Most questions are accompanied by foundational lectures (83.9%) and detailed explanations (90.5%), which provide general external knowledge and specific justifications for the correct answers. ScienceQA is the first large-scale multimodal dataset with annotated lectures and explanations for each answer. This dataset covers rich diversity across three domains: natural sciences, linguistic sciences, and social sciences, and is meticulously categorized based on topics, categories and skills.
数据集概述
数据集名称
- 名称: ScienceQA
数据集描述
- 概述: ScienceQA包含21,208个多模态多选科学问题,来源于小学和高中科学课程。其中10,332个问题(48.7%)包含图像上下文,10,220个问题(48.2%)包含文本上下文,6,532个问题(30.8%)同时包含两者。大多数问题附有基础讲座(83.9%)和详细解释(90.5%)。
- 特点: 数据集首次大规模注释了答案的讲座和解释,涵盖自然科学、语言科学和社会科学三个领域,具有26个主题、127个类别和379个技能。
数据集结构
- 数据字段:
image: 图像上下文question: 问题choices: 多选选项answer: 答案solution: 解决方案的思考过程CTH: 是否包含思考过程的标志
- 数据分割:
train: 16,966个样本test: 4,242个样本
任务与语言
- 任务类别:
- 问答
- 视觉问答
- 语言: 英语
数据集大小
- 大小类别: 100B<n<1T




