And as I have already said, I work more on splogs behaviour, rather than techniques. which make it easier to detect more techniques with really small algorythm.

I like the idea of marking unsure on ones you can't determine. Hopefully you can get that number down, but it is a really good start.

Your other problem is of course that the smarter spammers are going to adapt quickly. So your detection heuristics are going to have to constantly be adjusted.

Blogger will review manually to be sure they won't delete or take an action against a normal blog.

And yes you are right, currently the number of blogs analyzed is small, in the future a flag will be introduced to review manually the suspected results. Or Antisplog will just return a "Splogability rate" which will help to make difference between the 100% splog and the suspected ones.

I queries the database and currently there is less than 10% of suspected results. Until Antisplog can reduce it to less than 1%

For now the numbers of blogs you are dealing with may be small enough to keep an eye on, but once your crawler is working that isn't going to be possible. I know you are working to make the filter 100% accurate, but even experienced spam hunters have trouble determining the difference between some splogs and legitimate but strange blogs.

As you noticed there is some strange behaviours sometimes due to bugs or technique used to detect spam.

@Joe : Its the "keywords" section in your blog that have been detected spammy. You can use the contact link in the top, or the email feedback at antisplog.net

@Michael : www.recruiting.com and your blog www.specht.com.au are both concerned by the same bug, and its fixed now.

www.blogherald.com have a spammy behaviour even if it looks okay, that's why I'm using a safe list of keywords.

The same thing for www.identityblog.com but its a safe list of links.

I am interested in this service. I have some suggestions I would be interested in sharing privately (don't want to give all the secrets out to the spammers).