The Reuters-21578 dataset is a collection of documents with news articles. The original corpus has 10,369 documents and a vocabulary of 29,930 words. Source: Topic Model Based Multi-Label Classification from the Crowd