ScienceExamCER

ScienceExamCER是一个针对科学考试领域的高密度细粒度实体识别语料库，包含13.3万个实体提及。该数据集几乎所有内容词（96%）都被标注了细粒度的语义类别标签，如分类群组、部分-整体群组、动词/动作群组、属性和值、同义词等。这些标签来源于一个手动构建的包含601个类别的细粒度分类体系，该体系通过对4239个科学考试问题的数据驱动分析生成。数据集旨在支持科学领域中需要密集语义分类的问答下游任务。

ScienceExamCER is a high-density fine-grained entity recognition corpus tailored for the scientific examination domain, containing 133,000 entity mentions. Nearly 96% of its content words are annotated with fine-grained semantic category labels, including taxonomic groups, part-whole groups, verb/action groups, attributes and values, synonyms, and others. These labels are derived from a manually constructed fine-grained classification system consisting of 601 categories, which was generated through data-driven analysis of 4,239 scientific examination questions. This dataset is designed to support downstream question answering tasks that require dense semantic classification in the scientific domain.