From: Eivind Eklund Date: 2004-10-27T21:55:48+09:00 Subject: Re: Garden Spam On Wed, 27 Oct 2004 01:35:08 +0900, Brian Candler wrote: [ ... on using question based "captchas" with questions a human can recognize, but a computer cannot ...] > I can only think of a few different ways to generate the pool of questions > to show to the user and the expected answers: [ ... ways to get questions deleted ... ] > Others? The idea here is really using the ability of humans to recognize patterns and the set of facts that humans know to distinguish a human from a computer. An example of such a question is "Is the following page data from a legitimate page in this Wiki or from a spoof?", then showing a random page (or part of one) either from the legitimate Wiki pages in that Wiki, or from another Wiki. Unfortunate issue: This is possibly possible to overcome by mirroring the entire site and doing matching. I initially thought that that could be defended against by linking in the "fake data" and showing that as part of the real wiki (not much bother for visitors, noticable bother for bots), but unfortunately it may be possible to look at the interconnectedness of the two sets of pages (one from the real on-topic Wiki, the other from the "shadow" Wiki) and thus do a good guess on whether the page is real or fake. It DOES add significant hurdles, though. Eivind. -- Hazzle free packages for Ruby? RPA is available from http://www.rubyarchive.org/