[Wikiquality-l] Tagged revisions and trust coloring

[Luca de Alfaro]

Dear All,

I have been following the various threads on tagged revisions.
As many of you know, we have been developing a trust coloring of the text of
Wikipedia articles.  The trust value of a word depends on the computed
reputation of the author of the text, as well as on the reputations of all
visitors to the page; for more information, see
http://trust.cse.ucsc.edu/where a demo is also available.

In a short time, we will have  new (we hope better) version of the demo out,
and we will ask for your feedback; we plan by the end of October to have a
demo with all the English Wikipedia, as of the Feb 6 2007 dump (the last
complete one), colored.  We are still a few months away from doing the
coloring for an on-line, up-to-date, version of the Wikipedia, but we are
working towards that goal.

It may be fun to start thinking at how tagged revisions and trust coloring
may fit together.  My impression is that one would be able to get something
superior to either of them separately.  For  instance:

   - If an article lacks a manually tagged revision, one could select as
   "stable" revision a recent revision where as little text as possible is
   marked low trust (orange background).
   - Currenly, text becomes more trusted due to edits: if one edits a
   paragraph, one lends a bit of her/his own reputation to the paragraph (on
   the assumption that one reads the paragraph she/he is editing).  A mechanism
   similar to trusted revisions would enable users to say "I agree with this
   paragraph and I have checked it" without need for editing it.
   - If a manually tagged revision becomes quite old with respect to the
   most recent revision (we can measure age both in terms of edit distance and
   in terms of n. of intervening revisions), we could detect it, and offer
   instead of the automatically tagged revision, a revision that is more
   recent, and with no (or as little as possible) low-trust text.
   - We could monitor articles where the tagged revision becomes old (as
   above), and good candidate revisions are available later, and alert people
   in the "watch list" of the article, so that they might consider going to the
   article and tagging a newer revision.

These are just ideas that came on top of my mind; I am curious to know which
other suggestions or thoughts you might have.  As I said, we are still more
than a month away from a version of trust coloring that works for the
up-to-date wikipedia, but since we are thinking of how to architect the
system, it might be worth to think at how it fits with tagged revisions...

Another issue is this: might it be that, once the trust coloring is
available, the need for showing tagged revisions by default is lessened?  If
a trust coloring becomes only one click away, perhaps it is the
trust-colored and tagged revisions that should be available on demand, and
the up-to-date revision should be shown as default, as it has been so far?

I am very interested in your comments...
Best regards,

Luca
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.wikimedia.org/pipermail/wikiquality-l/attachments/20070927/6332e649/attachment-0001.htm