CROSSRE

CROSSRE是一个专为关系抽取设计的跨领域数据集，由哥本哈根信息技术大学和慕尼黑路德维希-马克西米利安大学合作创建。该数据集包含六个不同的文本领域：新闻、文学、自然科学、音乐、政治和人工智能，涵盖了超过5000个句子。CROSSRE不仅提供了多标签注释，还包含了注释过程中的元数据，如解释和困难实例标记。数据集旨在解决关系抽取模型在不同领域间的泛化能力问题，特别关注于模型如何从有限数据中学习并处理新领域中的新关系或实体类型。

CROSSRE is a cross-domain dataset specifically designed for relation extraction, co-created by the IT University of Copenhagen and Ludwig Maximilian University of Munich. The dataset covers six distinct text domains: news, literature, natural sciences, music, politics, and artificial intelligence, and contains over 5,000 sentences. In addition to providing multi-label annotations, CROSSRE also includes metadata from the annotation process, such as annotation explanations and markers for difficult instances. This dataset aims to address the problem of cross-domain generalization ability of relation extraction models, with a particular focus on how models can learn from limited data and handle novel relations or entity types in new domains.