If your use case for RSS/Atom feeds is to read mainstream news, my startup https://distilpost.com can fill in very well. News aggregated from over 1k sources as well as from tweets and instagram posts and categorized both by region and subject matter.
If you notice on the page, there are references to remarks made by actual people. The NER feature uses machine learning. Also the continuous clustering employs K-means and not just fuzzy matching as you stated. It is hard to imagine getting anywhere close to the level of accuracy with just fuzzy matching.
Pardon me if i am unfamiliar with the ways of the community but the down votes are simply far removed from what i expected as i only sougth to offer those deeply affected a possible alternative and something to cheer for.
Distilpost, my own startup is a more than adequate alternative. It is all machine curated at the moment and super accurate. Beats twitter moments easily.
https://distilpost.com/