Protect Me from What I Want

https://ontology2.com/essays/ClassifyingHackerNewsArticles/

although it definitely points to way to making an RSS reader which can learn to show you articles you find interesting. (You need some kind of clustering though to deal with avalanches of articles on the same topic...)

People go ahead making failing RSS readers with the same failing interface that failed in 1999 despite the fact that text classification technology was good enough in 1999 to have made something better. To be fair that prototype needs maybe 500 judgements to start getting good results but with dimensional reduction and clustering I think it could start looking relevant at around 50 judgements. One trouble is most people will give up in frustration after it doesn't learn after 5 judgements which might be possible if you think of it as a really hard game of "20 questions".