I love the idea, but I'm a little bit surprised that you manage to have an accurate prediction with so few data and so much parameters.<p>I've try once to do something like that, I've ended with something extremly simple. The score of a news was: a bonus for the number of likes, a malus for the number of days, and a bonus/malus per domain. The domain bonus was auto adjusted when I readed a news.<p>Nonetheless, the best function in this reader was to remove the links I had already open. This allowed, each time I browsed the page, to only have new links to explore.