BOXWRENCH
收藏资源简介:
BOXWRENCH是一个新的弱监督基准数据集,由威斯康星大学麦迪逊分校等机构的研究团队创建。该数据集包含五个文本分类任务,涵盖了高类别基数、类别不平衡和多语言变化等现实世界中的复杂场景。数据集包括Banking77、ChemProt、Claude9、MASSIVE18和MASSIVE60,分别用于在线银行查询、化学关系分类、不公平合同条款识别和多语言自然语言理解任务。数据集的设计遵循严格的标注函数(LF)设计流程,旨在为弱监督研究提供更真实的评估环境。该数据集的应用领域包括自然语言处理、化学信息学和法律文本分析,旨在解决弱监督在复杂任务中的实际应用问题。
BOXWRENCH is a novel weakly supervised benchmark dataset created by a research team from the University of Wisconsin-Madison and other institutions. The dataset encompasses five text classification tasks that address complex real-world scenarios such as high category cardinality, category imbalance, and multilingual variations. The dataset includes Banking77, ChemProt, Claude9, MASSIVE18, and MASSIVE60, which are respectively employed for online banking query, chemical relation classification, unfair contract clause identification, and multilingual natural language understanding tasks. Designed following a rigorous Label Function (LF) design process, the dataset aims to provide a more authentic evaluation environment for weakly supervised research. The application domains of this dataset include natural language processing, cheminformatics, and legal text analysis, with the intention of addressing practical application issues of weak supervision in complex tasks.

- 1Stronger Than You Think: Benchmarking Weak Supervision on Realistic Tasks威斯康星大学麦迪逊分校, 华盛顿大学, 斯坦福大学, 哈佛大学 · 2025年



