WALNUT
收藏资源简介:
WALNUT是一个专为自然语言理解任务设计的半弱监督学习基准。该数据集由微软研究院创建,包含8个不同类型的任务,涵盖文档级和令牌级预测任务。每个任务都包含由多个真实世界弱源生成的弱标签,以及一小部分干净标签。WALNUT旨在通过提供统一和系统的基准,促进弱监督在低资源自然语言理解任务中的研究和应用。数据集支持对不同弱监督方法和模型架构的有效性进行系统评估,并展示了弱监督在提升模型性能方面的潜力。
WALNUT is a semi-weakly supervised learning benchmark specifically designed for natural language understanding (NLU) tasks. Created by Microsoft Research, this dataset includes 8 distinct task types covering both document-level and token-level prediction tasks. Each task contains weak labels generated from multiple real-world weak supervision sources, along with a small subset of clean labels. WALNUT aims to facilitate research and application of weak supervision in low-resource natural language understanding tasks by providing a unified and systematic benchmark. The dataset enables systematic evaluation of the effectiveness of various weak supervision methods and model architectures, and demonstrates the potential of weak supervision in enhancing model performance.
WALNUT 数据集概述
概览
WALNUT 数据集是为 NAACL 2022 发表的论文《WALNUT: A Benchmark on Semi-weakly Supervised Learning for Natural Language Understanding》开发的基准数据集。详细的数据集描述和方法可以在这里找到。
数据获取
WALNUT 数据集可以从这里下载。
代码库结构
document-level-baselines:包含5种文档级分类任务的基线方法(C, W, Snorkel, C+W, C+Sonrkel)的源代码。document-level-GLC_MWNET_MLC:包含3种文档级分类任务的高级半弱监督学习方法(GLC, MetaWN, MLC)的源代码。token-level-baselines:包含5种词级别分类任务的基线方法(C, W, Snorkel, C+W, C+Sonrkel)的源代码。token-level-GLC_MWNET_MLC:包含3种词级别分类任务的高级半弱监督学习方法(GLC, MetaWN, MLC)的源代码。
引用
如果 WALNUT 数据集对您有用,请引用以下论文:
@inproceedings{zheng2022walnut, title={WALNUT: A Benchmark on Semi-weakly Supervised Learning for Natural Language Understanding}, author={Guoqing Zheng, Giannis Karamanolakis, Kai Shu, Ahmed Hassan Awadallah}, booktitle={Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, year={2022} }

- 1WALNUT: A Benchmark on Semi-weakly Supervised Learning for Natural Language Understanding微软研究院 · 2022年



