Open ghost opened 4 years ago
count could weight the similarity score more heavily etc, so constantly changing features (e.g. swapping computer or changing IP address) don't cause low score
You may also use ASN with or instead of IP (GeoLite2 ASN). Didn't see you mention this above.
Too privacy violating? I wanted to implement it for fun basically
If this command doesn't publish hashes, I see no problem with privacy.
Additional feature: display the zoom level and the real country code of the player.
This is about wight server but I didn't want to spam his repo with basically.. ideas
It's possible to fix up this impersonation / random nicknames situation with something I wanted to add as part of a separate 'stats' report for the old FFA.
Server knows lots of info about each client:
We don't want to violate privacy, but we could make it possible to figure out if say, player 'flimp' is similar to other previously seen players. Similarly, make it possible to figure out that 'detect' is not really detect.
Idea would just basically be taking a everything from above and splitting it out into a set of fixed features, (e.g. { HASH(ipv4), HASH(subnet), HASH(browsername), .. }) and providing a /whois command that computes jaccard similarity or similar over the target user vs all known users, and returning the top 3, or where a tie exists, the most frequently seen name.
Something like:
etc.
Too privacy violating? I wanted to implement it for fun basically
It would need some kind of small DB
(username, feature, count)
where count was maybe the total amount of time connectedcount
could weight the similarity score more heavily etc, so constantly changing features (e.g. swapping computer or changing IP address) don't cause low score