I’d like to just tell the intelligence trainer that I like or don’t like a post as a whole, and have it figure out which characteristics of the post I probably did or didn’t like.
For example, I might tend to dislike stories by a particular author on a multi-author blog, but I won’t necessarily notice that on my own, so I can’t use the fine-grained controls of the existing trainer very effectively.
This would have to feed into the rating system more gradually than the current trainer. One way to do it might be to use a 2-step process: 1) I rate posts to give the Trainer enough data to guess features; 2) the Trainer presents suggestions, which I accept, reject, or postpone, and the ones I accept drive the existing rating system.