Jimmy Wales wrote:
> Interesting stuff to look at. I wonder if we could find something like
> this useful at Wikia... especially at the larger sites.
Indeed and therein lies the problem. Developing an efficient algorithm
to computing something like trust, which relies on many complex and
interwoven variables/queries, requires great care. This care must be
emphasized, especially when users would require that the algorithm run
across the whole of Wikia in essentially real-time.
The key in allowing real-time computation of edit trust value is to
disconnect the author and their edits, at least temporarily.
Except for the initial edit, the trust value of words added by an author
should not affected by the future work of that author. They stand alone.
This disconnect makes some sacrifices in terms of temporal trust
resolution but because it can consider the current trust value of words
on a per article basis each article's word trust value can be computed
in an embarrassingly parallel fashion.
In order to enable author trust-base gain/loss we must update the trust
value of a particular author on a regular interval according to the
quantity and nature of the preserved edits. This can be accomplished
using standard edit histories combined with the previously calculated
word trust-value information. This approach also has the advantage of
being an embarrassingly parallel problem while minimizing trust-delta
network effects effectively limiting the scope of the trust-value update
to article->user without having to consider the newly updated trust
values of a very large number of other users.
Sacrifices required by the above approach include a reduction in
temporal user trust-base resolution in order to gain speed in computing
individual edit trust-values. Though this is not usually a problem so
long as an individual user does not edit enough times and with enough
quality to significantly change their entrenched trust-value -- not to
mention that we are concerned with an edits trust-value and only by
indirect association an authors trust-value.
This is not necessarily an ideal methodology, but it might address some
of the potential problems one might encounter when running such an
analysis on a system concurrently evolving through time.
> A lot of Wikia
> sites have very good high quality information but information that is
> generated without traditional encyclopedic sources. It seems like trust
> rankings like this could be useful in that context.
Yes, and even more so. We can extend this concept of trustworthiness in
a cross-wiki environment like Wikia. It opens up entirely new avenues
for existing trusted and new contributors.
It could allow trusted contributors to have an starting point of
authority when they reach out to new communities here at wikia and
elsewhere. If "trust" were differentiated by subject area knowledge it
might allow them to skip the whole "does this person know what they're
talking about" phase when they join a new wiki contributor community.
This technology could also do wonders for search engine ranking,
providing revision resolution, and a new way for people to search for
information. "Search most trusted", or "Search newest", and so on....
Another advantage to this new source of metadata is that a wiki could
hide untrusted edits by default ("show untrusted edits" off by default).
This would allow a sort of peer review (where the peers are the wiki
contributors who choose to show hidden revisions (available to
everyone)). This could potentially eliminate publicly visible vandalism
and spamming while maintaining openness. In addition, it could function
as a crowd-sourced anti-spam bot, as the crosswiki trust of the user
would approach zero through mass reversion of their edits, that user's
edits could be flagged for review as an example.
In short, I think it would be marvelous and a technological advantage to
have instantly available edit/user trust rankings on Wikia projects. Is
it possible? I would say it is. Is it a challenge in terms of both
development time, skill, and resource intensity? Certainly, yes. Do I
think we can do it? Most definitely. All we need is time and incentive,
and I think the potential for this technology is incentive enough.
_______________________________________________
Wikia-l mailing list
[email protected]
http://lists.wikia.com/mailman/listinfo/wikia-l
