textdetox/detoxification_pairwise_style_evaluation
收藏Hugging Face2025-07-22 更新2025-08-09 收录
下载链接：
https://hf-mirror.com/datasets/textdetox/detoxification_pairwise_style_evaluation
下载链接
链接失效反馈官方服务：
资源简介：
这个数据集旨在通过微调大型语言模型来评估去毒性效果，即生成的文本是否比原始文本毒性更低。数据集提供了两个文本的比较结果：原始句子毒性更大（去毒性效果良好），两者毒性相似（去毒性不足），或者生成文本毒性更大。数据来源于RUSSE 2022和TextDetox CLEF 2024任务中的人类注释，包含了多种语言的文本。
The dataset is aimed at fine-tuning large language models to assess detoxification quality - whether the generated text is less toxic than the original text. The dataset provides a comparison of two texts: the original sentence is more toxic (detoxification passed well), both sentences are similarly toxic (detoxification was not enough), or the generated text is more toxic. The data source is human annotations from the RUSSE 2022 and TextDetox CLEF 2024 tasks, including texts in multiple languages.
提供机构：
textdetox
