upb-nlp/EduMUSE

OpenStax多模态练习数据集是一个从OpenStax网站抓取的教科书练习题多模态数据集，每个练习都与其最相关的教科书子章节对齐，并在多个开放视觉语言模型下进行了评分。该数据集支持检索增强问答、视觉上下文对科学问答的贡献以及纯文本与多模态上下文的消融研究。数据集包含JSON文件和图像压缩包，结构按书籍、章节、练习嵌套，源内容基于CC BY 4.0许可。

The OpenStax Multimodal Exercise Dataset is a multimodal dataset of textbook practice exercises scraped from the OpenStax website. Each exercise is aligned with its most relevant textbook subsections and scored under multiple open vision-language models. This dataset supports retrieval-augmented question answering, research on the contribution of visual context to scientific question answering, and ablation studies comparing pure text contexts with multimodal contexts. The dataset contains JSON files and image archives, with a nested structure organized by book, chapter, and exercise. The source content is licensed under CC BY 4.0.