Neil Harris wrote:
I agree with Kai on this completely; if Wikipedia is about optimization by repeated random/evolutionary/whatever change, simulated annealing is _the_ classic way to achieve stable results.
Once an article had been progressively and slowly "cooled" to a low enough "temperature", it would be effectively frozen. If an article was shown to be seriously wrong, or needed extensive revision, it could always be "warmed up" again, either partially or all they way. The old "cooled" version of the article could be marked in the history as the "previous stable version".
In any case, as stable articles cooled down, they would change less and less often, making the use of the article rating process (where ratings must necessarily refer only to a single version) more and more useful.
Question: what would should a good algorithm for "cooling" and "heating" pages be based on? Article ratings for the last few versions? Consensus in an "articles for cooling" page? Intervention by admins?
Perhaps even some simple automatic heuristic like (for example) _very_ slowly cooling pages that are read repeatedly by a wide range of readers over some significant time period and yet not edited (ie, implicitly "validated" in a tiny way by those readers) during that time? Perhaps articles should slowly "heat up" if not read for some time?
I have suggested a statistically determined measure of an articles net rating based on a modified average of all recent individual ratings. The number of individual ratings upon which the value is based would also be noted. A low number of individual ratings would suggest an article in need of attention. This could be because the subject is so obscure that nobody ever pays attention to it, or the page has had so many recent edits that individual ratings have expired. If we add the total number of individual ratings to the data it will be obvious which of the two alternatives applies.
When statistical determinations are made it is important to remember the tendency of data to normalize itself. The individual vote becomes less important in its own right. The votes of trolls, POV pushers and other outlaws have a reduced statistical effect on the net result. I suppose that techniques could be built in to neutralize the effects of sockpuppetry if that is really a problem. Neutralizing the effects of sockpuppets is much better for the health of the community than acrimonious search and punish missions.
Ec