Hacker News Reader

Comment by BlueTemplar | Hacker News Reader

Yeah, you shouldn't count on it. Especially if they use neural networks - they likely have the know-how to identify you using your writing patterns (though the needle in the haystack issue might help you against not really persistent detectives...)

View on HN

"Especially if they use neural networks"

LOL.

"LOL" as in "heh, like they have the know-how to use them properly !" or as in "duh, of course they do!" ?

I think that was a LOL as in "your comment lacks understanding of the capabilities of neural networks in it's suggestion a neural network could identify someone in the pool of... everyone (lets even say in a certain area), based on writing style."

You'd need countless IDed writing samples to even give yourself a 50% chance of making a network able to with 50% accuracy match an unIDed writing sample to an IDed one.

Facial recognition is easier (yet still very hard) to implement because there're characteristics about people's faces that're hard to modify unless you're really trying.

Compared to writing style which could differ based on too many factors.

> In one experiment, researchers were able to identify 80% of users with a 5,000-word writing sample.

https://www.schneier.com/blog/archives/2013/01/identifying_p...

I'll just steal one of the comments from the article. "It isn't new, and it's just as overstated as previous studies. The 80% accurate identification rate applies only within the 5,000 subjects. The identification rate will plummet when the pool of writers expands to everyone who can write in English" Like I said, if you can narrow the pool to a small amount of people, than you probably could accomplish this (but even then, there's always the chance two people have similar writing styles.)

And if you had read further, you would have seen him corrected : it's 5k words, not 5k subjects. (And, also, I do raise this potential caveat in my mention of "needles"...)

Reply on news.ycombinator.com