AIPolished-Text Evaluation (APT-Eval)
收藏资源简介:
AIPolished-Text Evaluation (APT-Eval)数据集是由美国马里兰大学创建的,包含11785条样本,这些样本是从预先存在的人类撰写的样本中经过不同程度的AI润色得到的。数据集涵盖了六个不同的领域,旨在评估AI文本检测系统对AI润色文本的响应。该数据集可以用于检测AI生成的文本和人类撰写的文本之间的边界,以及解决因AI润色导致的误分类问题。
The AIPolished-Text Evaluation (APT-Eval) dataset was developed by the University of Maryland, United States. It comprises 11,785 samples derived from pre-existing human-written texts, which have been subjected to varying degrees of AI polishing. The dataset covers six distinct domains, and is designed to evaluate the responsiveness of AI text detection systems to AI-polished texts. Additionally, this dataset can be employed to discern the boundary between AI-generated and human-written text, as well as address the misclassification problems caused by AI polishing.

- 1Almost AI, Almost Human: The Challenge of Detecting AI-Polished Writing美国马里兰大学 · 2025年



