SciDMT
收藏资源简介:
SciDMT是一个大规模的科学提及检测语料库,由天普大学创建,旨在通过标注科学文献中的数据集、方法和任务提及,推动科学信息提取技术的发展。该数据集包含48,049篇科学文章,总计超过180万条弱标注提及,涵盖了广泛的科学领域和实体类型。创建过程中采用了远距离监督方法,通过与Papers with Code网站的文档级标注相结合,实现了高效的数据标注。SciDMT特别适用于科学论文索引、信息检索增强和科学知识的可访问性提升,为科学信息提取领域的模型开发和评估提供了强大的基准。
SciDMT is a large-scale scientific mention detection corpus developed by Temple University. It is designed to promote the advancement of scientific information extraction technologies by annotating mentions of datasets, methods and tasks in scientific literature. This corpus comprises 48,049 scientific articles, with a total of over 1.8 million weakly labeled mentions, spanning a broad spectrum of scientific disciplines and entity types. During its construction, distant supervision was employed, which was integrated with document-level annotations from the Papers with Code website to enable efficient data annotation. SciDMT is particularly well-suited for applications such as scientific paper indexing, information retrieval augmentation and enhancing the accessibility of scientific knowledge, serving as a robust benchmark for model development and evaluation in the domain of scientific information extraction.




