Re:2.1M of the oldest Usenet posts are now online for anyone...
by nl on Wednesday October 14, 2020 @01:52AM
(#24773532)
This really isn't how this works.
Stylometry is a thing, sure. But it's rarely accurate enough to use on bulk archives in an automated way, especially when archives come from different domains (eg, technical vs social writing).
It's useful for identifying possible alternate authors plays attributed to Shakespeare[1], but a lot less useful for trying to tie a reddit identity to a github profile.
Anyways, what can be done probabably will be done, for whichever reason because it's getting cheaper and cheaper.
Crawling and storing is cheap, but analysis is harder. Automated spell checking ("probabably" above not withstanding!) and auto-complete means stylometry is even less useful now.
Source: myself - I work in natural language processing, and have talked to people applying these techniques.
Stylometry is a thing, sure. But it's rarely accurate enough to use on bulk archives in an automated way, especially when archives come from different domains (eg, technical vs social writing).
It's useful for identifying possible alternate authors plays attributed to Shakespeare[1], but a lot less useful for trying to tie a reddit identity to a github profile.
Anyways, what can be done probabably will be done, for whichever reason because it's getting cheaper and cheaper.
Crawling and storing is cheap, but analysis is harder. Automated spell checking ("probabably" above not withstanding!) and auto-complete means stylometry is even less useful now.
Source: myself - I work in natural language processing, and have talked to people applying these techniques.
[1] https://www.frontiersin.org/articles/10.3389/fpsyg.2018.0028...