WikiBio GPT-3 Hallucination Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://huggingface.co/datasets/potsawee/wiki_bio_gpt3_hallucination
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是从维基百科传记中筛选出的子集，包含了长度排名前20%的238篇文章，用于评估文本嵌入属性。此外，该数据集被用于分析真实文本与生成文本的嵌入模式，特别关注生成文本中的幻觉场景。该数据集的规模为238篇文章，其任务是对文本嵌入进行分析，并对生成文本中的幻觉现象进行评估。
This dataset is a filtered subset sourced from Wikipedia biographies, comprising 238 articles ranked in the top 20% by their length. It is primarily utilized for two purposes: first, to evaluate text embedding properties; second, to analyze the embedding patterns of real and generated texts, with a special emphasis on hallucination scenarios in generated texts. With a total of 238 articles, this dataset supports core tasks including text embedding analysis and the evaluation of hallucination phenomena in generated texts.
提供机构：
Hugging Face
