InfoPop

该数据集名为InfoPop，是首个包含超过50,000份在线新闻文档中超过170万句子的流行度标签的数据集。它通过搜索引擎查询生成句子级别的流行度标注。在整理该数据集时，研究人员通过抓取新闻文档，并利用传入的搜索查询为单个句子分配流行度标签。此外，数据集还包含了一个清洗过程，以去除噪音和非语法句子。该数据集的规模涵盖了超过170万句子，来源于超过50,000份文档。其任务是对在线新闻文档中的句子流行度进行主动预测。

The dataset named InfoPop is the first of its kind to house over 1.7 million sentence-level popularity labels across more than 50,000 online news documents. Its sentence-level popularity annotations are generated via search engine queries: during the dataset curation process, researchers crawled news documents and assigned popularity labels to individual sentences using incoming search queries. Furthermore, a data cleaning procedure was conducted to eliminate noise and ungrammatical sentences. Spanning over 1.7 million sentences sourced from more than 50,000 documents, this dataset is designed for the task of actively predicting the popularity of sentences within online news documents.