So, I've been using the Defensio API lately. There's a "Defensio on Rails" plugin that I've had success implementing (with a few hacks specifically for the site using it, thank God for open source).
I've always been meaning to integrate Defensio into the anonymous blogging site I run, since I figured that sooner or later someone was going to target it for spam.
And I was right.
But the spam that eventually came wasn't your every-day automated "buy viagra now" spam. What the site got hit with was curious hybridization of trolling and spam; it was spam, no doubt, but it was clearly posted by actual people attempting to troll the site. Let's call the phenomenon "Troll Spam" for now.
After swatting away the first few posts by hand, I quickly looked into Akismet and Defensio's APIs. I was on a "short schedule" since I hadn't planned on spending my entire day implementing an anti-spam filter or doing the filtering by hand.
I eventually chose Defensio, and had it up and running in less than 2 hours. Why?
* Good API documentation: even though I was using the Defensio on Rails plugin (see above), I still needed to know the mechanics. Going through the Defensio API page was a breeze.
* RSS feed for "innocent" and "spam" pages. This was incredibly convenient for checking the results of the Defensio on the production site. I have my own little admin section that looks like it came out from a cracker-jack box, but Defensio's RSS feeds make it much better to monitor, real-time, what's ham and what's spam on the production site.
* Statistics. Tells me how the system is working. Not useful now, but fun to watch!
But by far the most important feature was that Defensio learns. The Troll Spam all had a similar theme, but different content. If you've ever seen a "raid" by 4chan or Something Awful goons, you'll know what I mean. Certain words, phrases, etc, repeated together. After training Defensio to HULK-SMASH the next few bursts of Troll Spam it's caught on to my intentions and started filtering out the bull-shit while still letting in the good-shit.
Meanwhile, the site continues to operate normally, the regulars not even realizing there's a Secret War happening!
Now, I know I could have spent 20 minutes creating my own filter that would have whacked posts that had these phrases in them, but this is so much better: Defensio has even started marking other posts by these trolls as spam as well. One of them thought they were going to be clever and posted an entire 3-page "ABC REPORTS" article. Never hit the front-page.
At the time of this writing they've basically just given up and started posting a long stream of obscenities (which is easier for them I guess), but none of it is getting through...
Statistically, Defensio has about a 50% accuracy, but it's been implemented less than 24 hours and its learning very, very quickly.
Smells like victory to me.
My only regret is that I didn't get Defensio implemented sooner -- I missed out on a lot of good learning data.