Reuters27000

To create the corpus, first we download from Reuters website 27,000 random news articles (HTML webpages) classified under each one of the following categories: Health, Art, Politics, Sports, Science, Technology, Economy, and Business. Next, we extracting from each article the title, the body and the category to which it belongs to. Finally, we stored in our database the title, body and category for each article that was downloaded. As a results—after removing duplicates—we obtained a corpus that comprises 23,863 documents that we randomly split in a training sequence that comprises 14,356 documents and a test sequence composed of 9,507 documents.

为构建该语料库，我们首先从路透社（Reuters）官网下载了27000篇随机选取的新闻稿件（HTML网页），这些稿件分别隶属于以下八大类别：健康、艺术、政治、体育、科学、技术、经济与商业。随后，我们从每篇稿件中提取其标题、正文以及所属类别。最后，我们将每篇已下载稿件的标题、正文与所属类别存储至数据库中。最终，在去除重复稿件后，我们得到了包含23863篇文档的语料库。我们将该语料库随机划分为包含14356篇文档的训练序列，以及包含9507篇文档的测试序列。