User talk:SvipBot/evaluation

I don't think I like this idea. - Quolnok 22:36, 18 November 2010 (CET)
 * At all? --Sviptalk 22:45, 18 November 2010 (CET)
 * No, not really. I'm sure the algorithm can be altered to recognise that edits like "Edit 74419 on "Countdown to Futurama": Score: 421 " are in fact good and possibly even identify edits that appear to be spam, but I don't think it will be able to pick up on grammar, spelling or, more importantly, whether someone is just making things up. It is also like the post count ranking systems on forums in that some people will try for good scores with useless edits (assuming this page is to be made more visible to general editors). - Quolnok 23:36, 18 November 2010 (CET)
 * Oh. Then you don't understand its purpose.  But that's my fault, I didn't clarify it on the page (I was in a rush).
 * Anyway, let me explain. I am working a system for the bot to 'revert' and 'patrol' edits.  The idea is, if the score is high enough (determined by a set value), the bot will revert the edit.  If the score is low enough, the bot will patrol the edit.  If neither of these are 'fulfilled', the bot will simple ignore the edit.
 * The purpose of this page is to test my algorithm, to see how the bot ranks these edits. See where it make mistakes and where it does right.  I am showcasing this to all of you, so that you can provide input.
 * When we have decided on a reasonable system, this page will not be updated any further. The idea is not to score people's edits for all to see, it's purely for testing purposes and decision making. --Sviptalk 00:00, 19 November 2010 (CET)

Bot actions
After tweaking the bot score system, I think I am finally ready to call for the actions to be taken by the bot based on score:


 * -inf to -1001: Mark as patrolled.
 * -1000 to 1000: Do nothing.
 * 1001 to 5000: Consider suspicious (or in short; don't perform clean up on this article)
 * 5001 to inf: Revert edit (unless already reverted of course)

Opinions before I launch this? --Sviptalk 03:13, 25 November 2010 (CET)
 * Hmm, does "don't perform clean up" mean it won't run the standard clean up, or won't revert/patrol? I'd vote the first option. "5001 to inf" there's been only one of these, right? Scores seem to be largely based on change in article size; What kind of score would creating an article the size of Bender's Big Score from scratch give? How about converting a similarly large article to a redirect? Can it tell if the edit was a page move? Does it ignore users who are Sysops? - Quolnok 02:05, 27 November 2010 (CET)
 * To answer your concerns/questions in order:
 * "don't perform clean up" means not to perform the general clean up and fixes that bot performs right now.
 * There are actually a few 5001 to inf right now, but only obvious ones (e.g. completely changing an article gives a very high score, around 10000, I think, unless it is a REDIRECT fix) get that high.
 * It is based a bit on size, yes, I am thinking of making it relative to the article size in the future, though. So it is only the change we are looking at in percentage.
 * Redirects edits are 'ignored' in regard to size.
 * It can tell if an edit was a page move, at present, it doesn't know how to handle such an edit (it will in the future).
 * It ignores sysops entirely. It will also ignore patrolled edits in the future as well.
 * I hope that answers some of your questions for now. --Sviptalk 09:32, 27 November 2010 (CET)