I wrote a free Twitter spam filter service, trained on thousands of accounts reported to @spam before the "report spam" button.

You'd enter your Twitter username, and asynchronously it'd grab your follower list and score each account for spamminess. Because of Twitter's rate limiting on the API, it could be a while before finishing the work if a lot of people were using the site at once. When done, it'd send you a @tweet with a link to get your results (and, if you want, click another button to block the spammy accounts).

Within an hour or two of people starting to use it, Twitter shut down the account. Tweeting the exact same message with a similar link at the end, requested by each and every user it tweeted to, was considered too spammy.

That was a year and a half ago. Nobody was reporting that account as spam, so they must have some basic heuristics already.