>>5767
yea this is why i don't post here that often anymore. i know the pissman has every poster memorized and will post their posting history if he gets mad enough. combine that with the fact that my vpn is now blocked since november...this is why smaller imageboards are uncomfy for me

https://people.eecs.berkeley.edu/~dawnsong/papers/2012%20On%20the%20Feasibility%20of%20Internet-Scale%20Author%20Identification.pdf

in 2012 they were able to identify the correct poster with 80% accuracy from a set of one hundred thousand posters basado on text corpus alone. and [b]classifiers are one of the few areas of technology that have exponentially increased in power since 2012[\b] and have a massive industry behind it.

of course a short post like on 4chan has a small amount of bits it can have and even things like uhh pidgeon hole principle can be used to say "wait a minute how many combinations of 7 word posts are there even!" (well its a lot actually) so theres no much information in one single post.... or is there? well yes there is! And in the example from 2012 that i skimmed it seems they were using blog posts which aren't usually that long anyway..


and its much harder to identify someone from 100,000 posters rather than from kissues' small population of what? like 100 posters? (maybe kissumins can shed some light on this, im also interested in any kind of analytics that they do collect)
also the difference between stylometry and fingerprinting is that any random person with access to kissu can implement a stylo attack and they dont need access to the server.
this is probably the 15th time I have made a post about stylometry on kissu, but also people keep making these threads about identifying people so I keep having to make them.

another cool fact about styolmetry attacks is that you can do totally cross platform attacks. for example finding someones 4chan posts, or other forum posts... or imagine the case where you have a large enough corpus built that you manage to find their facebook account! from the 2012 example using 100,000 people these seems kind of feasable.. Add together other bits of data you can find out about them. Usual posting time, descriptions about their life (gather demographic data)
using a few posts you might be able to find out exactly who someone is

Imagine the kind of abilities a place like the NSA has that is just a bunch of giant airplane hangars filled with nerds who love to think about this stuff all the time and have access to every single internet post and transaction in the entire world. It doesnt seem ridiculous to me that the NSA could take the text from one kissu post and then use that to get your SSN in a few hours if they wanted to.

very exciting!

>>5767
When it gets to the point where it becomes impossible to keep track of IPs is when analytic based moderation ends. I've already considered it, but it can be circumvented if you don't run javascript. Also I want to stop adding features after the year ends and fix administrative things rather than software things.

In the past two years or so no one's had a problem with the rules being based on removing people who are obnoxious, but it seems with more varied people that some would prefer an actual set of rules so they can point out discrepancies in policy.

If you want to talk about this kind of topic more, do it on /b/. It's not really suited for site meta to be on /qa/.

>>5768
If your VPN is banned send an appeal and it'll get the IP unblocked

>>5772
Make a post about site meta on /b/ in the future.

>>5774
There's a difference between questions about anonymity and asking why I haven't implemented fingerprinting

>>61442
just appeal a vpn ban, it'll be accepted without question (as long as the reason can be interpreted, saying 'hei' or something that confuses us will just get it ignored). only appeals that won't be accepted are tor appeals

that kinda sounds contradictory, let me clarify, if you say something like "my vpn" it'll probably be accepted and nobody has to know who you are

>>5767
Could be defeated if we all conspired to have the same fingerprint and ran a network of proxies so that we could post from each others' IPs. Although the latter might get abused. Maybe there could be cells of posters, each using a particular pool of IPs.

Of course, it would be better for everyone if Verniy just committed to not looking at IPs during meta arguments and maybe had IPs destroyed after a certain length of time.

i was thinking about ways to defeat stylometry
you could run your posts through a bunch of translators (would probably just make them more distinctive? randomize the translators? randomly use a computer thesorus to swap out words?) but i feel like that would actually make your posts more distinctive to a modern black box stylometry machine

perhaps a service that gets other anons to paraphrase your post for you?
that seems flawless, but severely impractical

googling someone else trying to express the same opinion as you and copy and paste their post? seems okay
can anyone think of something better?

>>5779
I recall hearing about some anti-stylometry software that would attempt to make your writing more average, but I think it was abandoned.

>>5778
>Maybe there could be cells of posters, each using a particular pool of IPs.
What is mobile posting?

>>5780
This actually sounds really doable. Formulas for evaluating how difficult (i.e. what grade level) something is to read have existed for decades now. Furthermore, word processors have been able to determine grammatical correctness for quite some time as well. Those features could easily be used to eliminate stylistic quirks, passive voice, and extraneous phrases. Also, word frequencies are fairly well known, which could give an approximate value for how "difficult" or "advanced" a given word is. That value could then be used to substitute words for easier, more common synonyms. In theory, then, it wouldn't merely be possible to make writing more average, but also allow for honing in at specific grade levels of writing, with the possibility of even making passages more advanced than they were originally written. To eliminate most attempts at stylometric tracking, it would probably be most preferred to randomize the writing difficulty. For further improvements, it would also be beneficial to determine the most common spelling and grammatical errors, as well as improper word usage (think, "should of" instead of "should have", or "irregardless" instead of "regardless"). For instance, then, you could reasonably transform a writing passage to look as if it was written on a smartphone by imitating auto-correct replacements.

It wouldn't surprise me if the various intelligence agencies already have a program as I've described.

>>5780
Found the software I was thinking about:
https://github.com/spencermwoo/anonymouth

Please don't give him any more shitty ideas...

>>5788
i have an intuition that this software would make your posts more distinctive because it would sound 'strange', and how modern stylometery is much more advanced than 'what word he used'
but if multiple people used it, then it would actually disguise you all as eachother

which is why kisumin should include an innovative feature that just automatically applies this feature to everyones posts and makes them all sound like shit
it would be even funnier if it reran with a different seed every time the page refreshed so each thread would look totally different to each reader

>>5790
but even then content and semantic analysis would still catch you...
there is simply no escape
if you're communicating over the internet you obviously have something unique about it. otherwise you wouldn't bother posting

probably a mathematically impossible to escape from identification as long as people are making novel posts.

>>5789
I'm not adding this.... It would create too many inconsistencies.

>>5792
Obviously not. It was just a jab.

>>5769
You talk about forensics as if it's a creepy technology thing, when really people have been able to do this sort of thing for over a century.