Good question. It's because a tag on one feed is not the same as a tag on another feed. TechCrunch will write about a company and I'll dislike it, but that doesn't mean I want to dislike that tag when John Gruber of daringfireball.net is writing about it.

It's really due to the fundamental nature of NewsBlur's filtering: it has to be explicit. NewsBlur will only filter what you tell it to. It's far easier to avoid false negatives (disliked stories showing up neutral) and false positives (liked stories showing up negative) by taking this approach. It does mean a bit more work for you, though.

One idea I've had is to have classifiers (tags/titles/authors) apply to an entire folder. But that's quite a bit of work, and may not even address the problem well enough.

I figured that might be the explanation. It's an old problem, and it is a really tricky one.

The only solution I've seen to this problem is from a failed platform about 5 years ago (http://rura.org/blog/2006/04/20/where...) -- basically, create semantic mappings between blogs to determine that tag A on blog X is equivalent to tag B on blog Y. Non-trivial. But when you get to the implicit rating system for Newsblur, you may have to do something similar anyway.

In any case, I'll just get back to rating everything for now.

Yup, implicit ratings are a planned future feature. The biggest issue is not the AI in the backend, but figuring out what a non-invasive yet intuitive UI would be for using and applying scores across feeds. It's critical that you know what NewsBlur is doing to score your stories, and that you have some way to correct for any behavior you don't want NewsBlur to do.