It's accurate enough that I had to create a new account now :)
I guess it's difficult to evade it as the word frequency certainly catches all about the countries I frequently refer, programming languages, interests etc.
Show HN: Using stylometry to find HN users with alternate accounts
241–250 of 562 posts
Re: Show HN: Using stylometry to find HN users with alternate accounts
#242On one hand, thank you for showing us all how easy it is to make something like this. No doubt organizations with more resources already have more sophisticated systems in the same vein. On the other hand, can we agree that this product is unethical? In many cases, when a person uses an alt, it is a direct and strong signal that they do not wish their other posts to be associated. So this product is circumventing the…
It does create a high level of discomfort, because it illustrates well what privacy advocates try talking about to the population at large, but all that said.. how is it any different from regular scraping and analyzing it any other way? This is a real question.
It's different because you're removing all barriers to access and making it easy and convenient to stalk/dox people.
Imagine you get the urge to track someone, but in order to do that you have to spend a week writing some new software. That's a barrier. And because of it you may change your mind because it's a lot of work with little payoff.
But if that info is just one click away, it's a whole different ballgame.
Re: Show HN: Using stylometry to find HN users with alternate accounts
#243Earlier quoted context omitted.
It’s a little writing trick I leaned from (I think) Orwell. Any time you’re about to use a common metaphor, try to tweak it. You’ll catch readers off guard, which piques their curiosity. It’s a fun game, too. I wish I’d used “the cat’s out of the hat,” but I didn’t think of it till later.
What you are describing is also known as an eggcorn. https://en.wikipedia.org/wiki/Eggcorn
Thank you! I was trying to find the original essay I learned it from. I’m now pretty sure it was by Poe, but all I can remember is the main advice: avoid common metaphors.
I vaguely remember one of the metaphors in the essay was about a chicken coop melting, or something like that. It was vivid enough to leave a big impression.
Re: Show HN: Using stylometry to find HN users with alternate accounts
#244After a few tries on boring accounts, I thought to try the account of somebody who was notorious for an incident outside of HN, and had a (deservedly) bad time at HN for a couple of years before the account went dark. And yeah, there's a bunch of high confidence (.6-.8) hits for that account, and from a quick browse of the comments of the recently active ones, they look really likely to be alts. Like, all three that…
MD5 of the username is 9abc27e93b7e3c04b7c599017c1cfe5f ? The top one seems an odd one out in that case?
Re: Show HN: Using stylometry to find HN users with alternate accounts
#245This found an alt that I created specifically to see if I could write artificially to defeat this kind of analysis. I have seen other tools like it posted to HN, but none before had found that account. I guess I need to up my game.
If you don't mind sharing, are you "writing artificially" purely in your head, or are you using techniques like intermediate translations?
Re: Show HN: Using stylometry to find HN users with alternate accounts
#246Found my phone account; I'm quite impressed, really !
Re: Show HN: Using stylometry to find HN users with alternate accounts
#247I have two accounts. This one, “soneca”, that is my first one and most active by far, and another one that I use sometimes mostly for Show HN and few comments.
When I searched the other one, “soneca” was the first guess, with 0.4.
But when I searched “soneca”, the other one was not in the top 20.
Re: Show HN: Using stylometry to find HN users with alternate accounts
#248Earlier quoted context omitted.
False positives become an increasingly difficult problem the more and more potential authors you introduce. If I had wrote a fancier model it probably wouldn't be as much of a problem but what can you do.
Yes, this wasn't a criticism of the tool. It is crazy good. But I don't think people should be making the assumption that bolded results are definite alts, which sillysaurus' comment reads like.
Hmm, that wasn’t my intent. I see this tool as a recommendation engine more than a doxxer. By “signal to noise ratio close to infinity,” I meant that if you visit one of the bolded accounts, they’ll probably sound a lot like you.
It’s one of those ideas that makes the tool substantially more effective, yet never would’ve occurred to me. It’s like the simplicity of pg’s “a plan for spam” algorithm: deceptively simple, but (like scrubbing dishes with fingers) works really well.
Re: Show HN: Using stylometry to find HN users with alternate accounts
#249What algorithm is being used?
Re: Show HN: Using stylometry to find HN users with alternate accounts
#250Earlier quoted context omitted.
The explanation is here: https://news.ycombinator.com/item?id=33755466 As far as I’m concerned, it’s the killer feature of the app. The top 20 results may be noisy, but the bolded results have a signal to noise ratio close to infinity.
I have 7 bolded names (0.53-0.62) in the top 20 list, and none are alts of mine.
Pretty much the exact same. (I do have a throwaway account but I rarely use it and it probably hasn't been used enough to qualify.)