I've been thinking about a very similar idea. Mostly so I don't feel the compulsion to check the internet so frequently, the good stuff would just wait until I do.
Would searching through the text be enough though? If you could get several people to use it that would give it more information. You could rank content based on whether or not someone else with similar interests has liked it on top of that.
There are various things you can do and various machine learning techniques you can use, but I imagine that the single user version would be enough, to start with.
I'll give it a go and see if it works well. If it does, I might release it as a service.
That would be cool. The basic info might be enough. The thing about the previous guy who tried something like this which was posted above is, he was mostly sorting by the words in the title which seems like only a very weak predictor. Better would be the number of comments and votes, and maybe other stuff like how long the article is or whether or not certain words are in the comments. He also trained it on whether he thought it sounded interesting, not after reading the article and determining if it actually was.
I plan to implement votes (maybe), domain, actual raw text of the article, title, submitter (maybe) and show articles that are deemed "important" (i.e. have stayed on the front page for longer than X hours), as well as some random ones, to avoid a bubble. Plus, I've already started training the filter manually, I'll maybe write a simple web UI later on so I can up/downvote articles from there.
Would searching through the text be enough though? If you could get several people to use it that would give it more information. You could rank content based on whether or not someone else with similar interests has liked it on top of that.