Reuters 21578 dataset in json and sgm format, and the conversion script.
Uses BeautifulSoup for XML parsing:
pip install BeautifulSoup
The entire original data can be found in other-files and sgm-data. You can find the original archive on archive.ics.uci.edu