troll detection algorithms

Post Reply New Topic

okay now you got me

jk

it's a hobby

i don't know why you would even glorify linus anyway. what he allegedly did is impossible, therefore he's an impostor

http://www.ipvoid.com


Something like that utilizes blacklists of known bad IPs. Large lists like these can be used to prevent people on known VPN addresses (which is most of them pretty quick as there are only a limited amount), known malicious addresses, known hacker or scammer addresses, etc. from accessing whatever by simply being placed on a blacklist.

well that's just the basic stuff though (which looks like this site isn't even implementing yet. i don't know if it's just that alex is too busy with other things or if there's actually some kind of specific reason for him to be avoiding more reliable automatic spam detection, for instance)

those tools and techniques can be improved and they should certainly be used in conjunction with other techniques no matter what, and keeping their usefulness in mind is always important during both development and implementation (no need to reinvent the wheel). but still, they're not my focus

thanks for the link anyway

Well I am not sure who runs what but with the combination of cloudflare and proper blacklisting techniques there is no better option without banning IP ranges of innocent people. If all the VPNs are banned and malicious IP addresses then each person only has one IP or a very limited number of IPs to get around (so just keep banning, they will eventually run out). The whole reason for the switch from ipv4 to ipv6 is because with ipv4 there are only a limited number of total IP addresses and we need more to be available. If anyone cannot get in a simple e-mail to an admin or contact should suffice for whitelisting purposes. It's not a troll detector but it prevents the most trolls by making them only be able to rely on one or a very limited number of IP addresses.

to be honest i don't have the exact real-life applications figured out yet, just a thought so far. and i don't know either if i am ever going to build an actual troll-spotting system. i don't really have a personal investment in the solution, i'm more interested in the problem itself. i haven't been coding much lately, and this seemed like the perfect excuse for a project-for-the-sake-of-a-project, because it will be fun no matter what, and it will motivate me to learn more about the topics i find the most interesting

the "end result" will probably be just yet another completely unrelated app on an app store and, who knows, maybe a few bucks . i just want to shake away the rust

You could try Spamsum and couple it with a classification tool. Spamsum is a fuzzy hashing scheme (the similarity between this and ordinary hashing can be seen as analogous to the difference between the mean of a function and the running mean).