It’d also be good to have an output RSS feed, containing javascript tick and cross buttons, so we could read the output elsewhere.

Re: your requests:

1) We’re experimenting with a few options here I believe.

2) Bookmarking – definitely something I want to see. Maybe something similar to Google Reader’s star system would do?

3) That is a cool idea; maybe some sort of ‘tag cloud’ kind of thing that shows the weightings of terms you like/don’t like.

Re: algorithm: it seems to work pretty well so far and afaik we don’t have anyone that is not getting any items in their filter. The key thing for the algorithm appears to be maintaining a good balance between items you’ve trained as ‘good’ and items you’ve trained as ‘bad’. One of the things on our TODO is some sort of visual indicator to help you get this balance right.

In my case I use (or intend to) feedzero to filter through the new scientific literature that comes out in my field. As that is about 50 papers / abstracts per day, having a somewhat brainy RSS reader that narrows that down is a blessing. Looking forward to the improvements .

requests thus far:
* some kind of settings dialog, where i can alter my settings for hiding / showing things I marked as disliked, things I probably dislike, things I marked as liked etc.
* some kind of bookmarking system so that I can store certain entries for later browsing.
* for fun: some (wordle?) representation of what your bayesian filter has learned so far.

suggestion for the algorithm:
I am not too familiar with spam trainers, but it might be worthwhile to let a user-settable percentage of the feeds slip through without being weighted. In this way you might give people that have overtrained their filter into not presenting anything with a few particular keywords into adding some nuances.