Anthere wrote in part:
Having far less participants, non-english wikipedias
do not "work" 'round the clock. There are periods of
time when nobody is watching, and nobody can react to
an attack. To speak of the case I know, there is very
often noone between 2am and 5am, and sometimes noone
between 9am to 12am on the fr:wiki. Or when somebody
is around, it's often a newcomer.
If any vandalism was to occur from an anonymous ip, we
are only 3 sysops to be able to ban it anyway.
I support system that could automatically detect a
potential problem.
It is very likely that
- one user/ip saving every minute for more than 10 mn
is a potential problem
- any edit replacing more than xxxx characters by x
characters (except redirect...) is a potential problem
- any edit replacing more than xxxx characters by an
image that was downloaded less than tt minutes before
is a potential problem
But, then, what good would it do, if an automatic
system detect a potential problem but has no human to
warn ?
OK, here's an idea:
We should develop (on [[m:]]) a Bayesian vandal detector.
We'll implement this detector (or versions of it) on every wiki,
and set up a mailing list of militia members to be warned
when the detector suspects vandalism.
The catch is, the mailing list is international.
So if there's vandalism on [[fr:]],
then I can learn about it and respond to it
during the time that I'm online,
even if none of the French speakers are online then.
(Of course, I need to know a little French to do this,
so I can list the languages that I know a little of
when I sign up for the mailing list,
lest I get warnings about vandalism on [[zh:]]
that I can't safely do anything about.)
-- Toby