NoisyAG-News
收藏资源简介:
NoisyAG-News是由杭州电子科技大学开发的文本分类基准数据集,旨在评估实例依赖噪声学习方法的有效性。该数据集包含50,000条通过非专家人工众包注释的样本,涵盖四个类别。数据集的创建过程包括从AG-News数据集中选择样本,并通过多轮注释和质量控制确保数据质量。NoisyAG-News主要用于解决文本分类中的实例依赖噪声问题,特别是在预训练语言模型和噪声处理技术中的应用。
NoisyAG-News is a text classification benchmark dataset developed by Hangzhou Dianzi University, designed to evaluate the effectiveness of instance-dependent noise learning methods. This dataset contains 50,000 samples annotated by non-expert human crowdworkers, covering four categories. The dataset was constructed by selecting samples from the original AG-News dataset, and ensuring data quality through multiple rounds of annotation and quality control. NoisyAG-News is primarily used to address the instance-dependent noise problem in text classification, especially for applications involving pre-trained language models and noise handling techniques.

- 1NoisyAG-News: A Benchmark for Addressing Instance-Dependent Noise in Text Classification杭州电子科技大学 · 2024年



