At 06:22 PM 2/4/2004, you wrote:
Guillaume Blanchard wrote:
There are many other criteria that can define
"questionable modifications",
but this can handle a lot of case we run up against. Those questionable
modifications may be displayed in RC (for sysop or for all) with a
different
mark (color, bold, italic or any visible marker).
The goal is to give
Could you use SpamAssassin for this? It assignes a probability that a
message is a spam, based on how previous messages were categorized.
Filtering based on
http://en.wikipedia.org/wiki/Bayes%27_theorem
or rather
http://en.wikipedia.org/wiki/Bayesian_inference
Can we dig out cases of vandalism from the "old" database table, and
use them as training fodder for SpamAssassin?
SpamAssassin may indeed use Bayesian filtering but is primarily a heuristic
(rule based) engine. I believe you could create a heuristic engine that
would mark "likely" vandalism on the Wiki, however, just like spam, it only
starts an arms race. The spammers know about SpamAssassin, so they do
things to get around it. The next version of SpamAssassin deals with these
tricks (like word salad, for example). Then the spammers do something more
tricky. and so forth.
In the world of the Wiki, there is no economic advantage to vandalism (that
I can think of anyway) so perhaps it would not be as serious an escalation
as there is in the world of spam.
Eventually, I'm afraid that you will have to REQUIRE logging in with a user
name prior to allowing edits on the Wiki. At least that way you can undo
all changes by user XXX automatically...
-Kelly