[brohee]

Keyword matching doesn't work, never did without way too much collateral damage. Being able to filter "I will rape you tonight" based on the word "rape" is of no use at all to someone that actually wants to discuss the subject of rape, but doesn't want to be threatened by it.

This is a social problem without a technical solution, even not permitting blocked people of following and retweet like she'd like to is of litle use when it takes seconds to create a new account...

On the bright side, I don't think there is much in the way of following up on those threats in meatspace, cowards will be cowards...

[drb311]

Is identifying a threatening or abusive tweet any harder than identifying a spam email? Modern filters can get rid of most spam mails pretty well, and don't delete much that the receiver will miss.

[brohee]

If your legitimate email dealt a lot with replica watches and erectile dysfunction medication, your spam would happen to look a lot more like what you need to see, and I'm pretty sure you'd be unhappy with the current antispam technology...

The issue here is that the offensive text is so close to the desired text, to the point that quoting threatening material makes it non threatening. The tweet 'He said "You will be raped tonight"' may be as non threatening as it gets, while "You will be raped tonight" should land its author before a judge.

I'm not saying it's impossible, but it's IMHO very hard.

[drb311]

Yes, that does make it harder.

But a few things make it easier too:

1. Twitter can see all messages from the sender -- if the stream contains a lot of suspect material and was created recently (for example), that makes any individual Tweet more suspect. No individual mail client can do this. 2. A 'false positive' in Twitter has a lower cost than in email. People ignore tweets all the time, so hiding the wrong thing sometimes won't matter so much. 3. Twitter is a more constrained platform, so the problem domain is presumably more limited.

I don't mean to say it's easy or even "not very hard" to pull this off -- just that it is either well within Twitter's capabilities, or soon could be.