Although IP blocking isn't perfect, it's probably the most practical. How much worse would spam be without the current blacklists? Although machine learning is always an alluring route to take, it is very very very hard to get right and is still easily tricked. 99% of the mail in my Yahoo! spam folder is not spam. But, if you have a unique way to better detect spam using machine learning that doesn't require constant review, then I would gladly pay you for that service.
________________________________ From: Daniel Friesen <[email protected]> To: [email protected] Sent: Friday, May 24, 2013 7:09 PM Subject: Re: [MediaWiki-l] Wiki spam. Stronger fightback. On Fri, 24 May 2013 13:41:04 -0700, Al Johnson <[email protected]> wrote: > > Maybe mediawiki sites can unite to keep a global list of these IP's and > block them as soon as they are submitted. Each mediawiki site can > auto-submit a spammer IP as soon as it's discovered to the global list. > What are the problems with this idea? > > Al IP blocking simply doesn't work. It's like playing whack-a-mole against a billion moles (or trillions on trillions once IPv6 really takes off). There are too many open proxies, botnet machines, etc... and many of them are either also addresses used by real editors, NAT addresses with editors on them, or dynamic IPs that will soon be forced on a non-spammer while the spammer gets an unblocked IP. The proper way to deal with this spam is not by IP but by content. We need some people who are knowledgeable about matching spam by training programs with spam and non-spam. That's the kind of central database that would be useful. An extension that sends spam (and after awhile things marked non-spam) to a central database. A community on that database that vets valid and invalid submissions. And eventually a mode for that extension that will start using information generated from that data to start filtering out spam edits. I've actually already thought about this and thought about how to make it friendly to users when their edits accidentally end up considered spam: https://www.mediawiki.org/wiki/User:Dantman/Anti-spam_system -- ~Daniel Friesen (Dantman, Nadir-Seen-Fire) [http://danielfriesen.name/] _______________________________________________ MediaWiki-l mailing list [email protected] https://lists.wikimedia.org/mailman/listinfo/mediawiki-l _______________________________________________ MediaWiki-l mailing list [email protected] https://lists.wikimedia.org/mailman/listinfo/mediawiki-l
