KPTimes

KPTimes是一个大规模的新闻文档关键短语生成数据集，由法国南特大学的LS2N创建。该数据集包含279,923篇新闻文章，每篇文章都配有编辑精心挑选的关键短语。数据集的创建过程涉及从选定的新闻网站爬取文章，并使用HTML元数据中的关键短语作为黄金标准。KPTimes旨在解决现有数据集主要针对学术领域且包含非专家注释的问题，特别适用于训练和评估神经关键短语生成模型，尤其是在新闻领域中的应用。

KPTimes is a large-scale keyphrase generation dataset for news documents, developed by LS2N from the University of Nantes, France. It contains 279,923 news articles, each paired with editor-curated keyphrases. The dataset is constructed by crawling articles from selected news websites, and using the keyphrases embedded in HTML metadata as the gold standard. KPTimes aims to address the limitation that most existing datasets are restricted to the academic domain and utilize non-expert annotations, and is particularly suitable for training and evaluating neural keyphrase generation models, especially for news-domain applications.