That's one interpretation of the confusion matrices on page 4, here: http://upload.wikimedia.org/wikipedia/wikimania2007/d/d3/RassbachPincockMingus07.pdf

44% of FAs were classified correctly, but 17% were classified as A and 20% as good.
37% of As were classified correctly, but 15% were classified as FA and 14% as G
41% of Gs were classified correctly, but 20% were classified as FA, 14% as A and 10% as B

Collapsing over these greatly improved classifier performance. One interpretation, which is partially true, is that not all of the features necessary to predict article quality were included. But another, which is also partially true, is that the quality classes used are artificial, and that not all articles are classified correctly.

Nonetheless, I provided featured articles as an example because documentation of their loss in quality is documented, whereas in other classes it is not. And my own opinion is that the featured article process does promote quality. They are fairly easy for the classifier to pick out in most cases. Whether or not there is a large swath of articles which deserve FA status which do not have it is only up to speculation. We will have to wait for those articles to go through the process. It's certainly true that there are some hidden nuggets.

On 8/30/07, Kat Walsh <kat@mindspillage.org> wrote:
On 8/29/07, Brian <Brian.Mingus@colorado.edu> wrote:
> I am thinking more along the lines of the loss of quality of previously high
> quality articles, which are already incredibly small in proportion, such as
> "featured articles." Traditional content production methods asymptote in
> quality, but the editing process in place at Wikipedia (which is only one
> possible wiki process, and also one of the most successful, but does not
> necessarily speak about wikis in general) encourages articles to gradually
> increase in quality, and then again decrease. It is unknown if they will
> stabilize (which brings about thoughts of a 1.0)
>
> There are plenty of examples of this phenomenon:
> http://en.wikipedia.org/wiki/Wikipedia:Former_featured_articles
>
> This could be due to changing featured article criteria, but in general, the
> claim that simply starting a wiki encourages high quality content is lacking
> evidence. If anything, wikis encourage the addition of noise to high quality
> content. Adding noise to turing complete wiki syntax can quickly snowball,
> turning into an aggregation of media that lacks coherence.

I think that monitoring featured articles is a poor metric for quality
of Wikipedia articles in general, as getting an article "featured" on
English Wikipedia is not solely a function of article quality -- some
poor work gets through and some good work will never get through
(because the subject doesn't merit a long article, for example), or in
some cases those writing articles have no desire to put them through
the featured article process.

The ratings by the individual wikiprojects, while still wildly
variable and idiosyncratic, are probably a better guide.

-Kat

--
Wikimedia needs you: http://wikimediafoundation.org/wiki/Fundraising
* *  * *  * *  * *  * *  * *  * *  * *  * *  * *  * *  * *  * *  * *
http://en.wikipedia.org/wiki/User:Mindspillage | (G)AIM:Mindspillage
mindspillage or mind|wandering on irc.freenode.net | email for phone

_______________________________________________
Wiki-research-l mailing list
Wiki-research-l@lists.wikimedia.org
http://lists.wikimedia.org/mailman/listinfo/wiki-research-l