A team led by Rebecca Portnoff has used AI/data mining techniques to identify sex traffickers on Backpage. https://www.newscientist.com/articl...rail-to-find-and-help-sextrafficking-victims/ https://dl.acm.org/citation.cfm?doid=3097983.3098082 The tool developed & used by the researchers in combination: identified the style in which ads are written, using AI searched the bitcoin blockchain to identify the wallet associated associated the timestamp of the ad posting It correctly identified about 90% of ads that had the same author, with a false positive rate of only 1%. The relatively new aspect of this is the use of AI to characterise "style", and doing so in conjunction with more traditional cues. I've written in the past about how individual use of pronouns is diagnostic, I imagine (there are no further details), that the way this AI works is more general. Sadly, this class of technique is increasingly likely to be used in less laudable de-anonomisation attacks. This might give people pause for thought when taking the necessary care over opsec and segregation - as Mirimir has posted.
Yes, this is scary stuff. My meatspace persona does very little online in English. Mirimir, in case you haven't noticed, is a kludge of various variants of English that I've been exposed to. Plus some Mexican Spanish. And other personas have used other languages than I know well enough. But there's a limit to that, of course. I wonder how well those AI approaches work across languages.
I am guessing writing style is pretty idiosyncratic and differing languages only slow the process of determination down a bit.
Well, there is no "ain't" either But yes, point taken I especially love "all y'all". I mean, one can "every one of you", but this is so much shorter. And "some y'all", that's just awesome.
Gorgeous! I guess there's all the language/regional/dialect variants, as well as characteristic use of slang/idiom/phrases, being quite diagnostic - let alone in-group slang. For example, there's a group of people who will know the phrase about humans - "mostly harmless" - from Douglas Adams. I do like the idea of running text through multiple translations, though I guess that would have to avoid cloud services.
If you look in junk shops -- or in my case, junk boxes -- you can find pre-Internet translation software. With local databases. You may need to run an XP (or even 95) VM, but hey