> the end result is going to have countless false positives

I suspect the real measured false positive rate is remarkably low. Cases like this make great headlines but still impact only tiny fraction of Twitters hundreds of millions of users and happen rarely compared to the message volume.