Man, things like that have been clamored for by the curating community for a long time (and regular folks as well). I can't say how easy or hard it would be to train something to 'find' good content, but as a curator I know in the first 10 seconds if something is going to be worth reading or not. At the very least, it'd be nice to have the chaff filtered out, leaving you with things that at least look like wheat.

Build it, and let me know when you have it done :)