I've long considered using bayesian classification to mark feeds for like/dislike so the system can learn what I like, but from what I hear, the amount of text

I've long considered using bayesian classification to mark feeds for like/dislike so the system can learn what I like, but from what I hear, the amount of text in the feed may not be sufficient to classify properly. I have previously looked at http://uclassify.com http://uclassify.com that provides such a web-service that classifies text based on various criteria. I also remember signing up for an actual rss reader web-service that planned to use bayesian classification to learn your likings a couple of years, but they never released -- I don't recall the site name right now.