Hoi,
When you have a system that reports on what needs a simple response you do
not report, you add a lable. It is the lack of such considerations why it
is a Wikipedia approach and not a Wikidata approach. The tool will rate
items and it will be largely meaningless.
When the idea is that we have at least something.. Fine. But do not mistake
it for quality, meaningful quality.
Thanks,
GerardM
Op wo 22 mrt. 2017 om 18:05 schreef Amir Ladsgroup <ladsgroup(a)gmail.com>
I was mentioned as "the developer of ORES".
So I comment on that. Aaron
Halfaker is the creator of ORES. It's been his work night and day for a
few years now. I've contributed around 20% of the code base. But let's be
clear, ORES is his brainchild. There is an army of other developers who
have contributed. E.g. He7d3r, Jonas.agx, Aetilley, Danilo, Yuvipanda,
Awight, Kenrick95, NealMCB, and countless translators. The idea that a
single person can develop something like a production machine learning
service. Yikes.
See:
https://github.com/wiki-ai/revscoring/graphs/contributors (the modeling
library)
https://github.com/wiki-ai/ores/graphs/contributors (the hosting service)
https://github.com/wiki-ai/ores-wmflabs-deploy/graphs/contributors (our
server configuration)
https://github.com/wiki-ai/wikilabels/graphs/contributors (the labeling
system)
https://github.com/wiki-ai/editquality/graphs/contributors (the set of
damage/vandalism detection models)
https://github.com/wikimedia/mediawiki-extensions-ORES/graphs/contributors
(mediawiki extension that highlights based on ORES predictions)
Also, I fail to see the relation of running a labeling script to what's
ORES is doing.
Best
On Wed, Mar 22, 2017 at 8:51 PM John Erling Blad <jeblad(a)gmail.com> wrote:
Only using sitelinks as a weak indication of quality seems correct to me.
Also the idea that some languages are more important than other, and some
large languages are more important than other. I would really like it if
the reasoning behind the classes and the features could be spelled out.
I have serious issues with the ORES training sets, but that is another
discussion. ;/ (There is a lot of similar bot edits in the sets, and that
will train a bot-detector, which is not what we need! Grumpf…)
On Wed, Mar 22, 2017 at 3:33 PM, Aaron Halfaker <aaron.halfaker(a)gmail.com>
wrote:
Hey wiki-research-l folks,
Gerard didn't actually link you to the quality criteria he takes issue
with. See
https://www.wikidata.org/wiki/Wikidata:Item_quality I think
Gerard's argument basically boils down to Wikidata != Wikipedia, but it's
unclear how that is relevant to the goal of measuring the quality of
items. This is something I've been talking to Lydia about for a long
time. It's been great for the few Wikis where we have models deployed in
ORES[1] (English, French, and Russian Wikipedia). So we'd like to have the
same for Wikidata. As Lydia said, we do all sorts of fascinating things
with a model like this. Honestly, I think the criteria is coming together
quite nicely and we're just starting a pilot labeling campaign to work
through a set of issues before starting the primary labeling drive.
1.
https://ores.wikimedia.org
-Aaron
On Wed, Mar 22, 2017 at 6:39 AM, Gerard Meijssen <
gerard.meijssen(a)gmail.com> wrote:
Hoi,
What I have read is that it will be individual items that are graded. That
is not what helps you determine what items are lacking in something. When
you want to determine if something is lacking you need a relational
approach. When you approach a award like this one [1], it was added to make
the award for a person [2] more complete. No real importance is given to
this award, just a few more people were added because they are part of a
group that gets more attention from me [3]. For yet another award [4], I
added all the people who received the award because I was told by someone's
expert opinion that they were all notable (in the Wikipedia sense of the
word). I added several of these people in Wikidata. Arguably, the Wikidata
the quality for the item for the award is great but it has no article
associated to it in Wikipedia but that has nothing to do with the quality
of the information it provides. It is easy and obvious to recognise in one
level deeper that quality issues arise; the info for several people is
meagre at best.You cannot deny their relevance though; removing them
destroys the quality for the award.
The point is that in relations you can describe quality, in the grading
that is proposed there is nothing really that is actionable.
When you add links to the mix, these same links have no bearing on the
quality of the Wikidata item. Why would it? Links only become interesting
when you compare the statements in Wikidata with the links to other
articles in the same Wikipedia. This is not what this approach brings.
Really, how will the grades to items make a difference. How will it help us
understand that "items relating to railroads are lacking"? It does not.
When you want to have indicators for quality; here is one.. an author (and
its subclasses) should have a VIAF identifier. An artist with objects in
the Getty Museum should have an ULAN number. The lack of such information
is actionable. The number of interwiki links is not, the number of
statements are not and even references are not that convincing.
Thanks,
GerardM
[1]
https://tools.wmflabs.org/reasonator/?&q=29000734
[2]
https://tools.wmflabs.org/reasonator/?&q=7315382
[3]
https://tools.wmflabs.org/reasonator/?&q=3308284
[4]
https://tools.wmflabs.org/reasonator/?&q=28934266
On 22 March 2017 at 11:56, Lydia Pintscher <lydia.pintscher(a)wikimedia.de>
wrote:
On Wed, Mar 22, 2017 at 10:03 AM, Gerard
Meijssen
<gerard.meijssen(a)gmail.com> wrote:
In your reply I find little argument why this
approach is useful. I do
not
> find a result that is actionable. There is little point to this
approach
and it
does not fit with well with much of the Wikidata practice.
Gerard, the outcome will be very actionable. We will have the
groundwork needed to identify individual items and sets of items that
need improvement. If it for example turns out that our items related
to railroads are particularly lacking then that is something we can
concentrate on if we so chose. We can do editathons, data
partnerships, quality drives and and and.
Cheers
Lydia
--
Lydia Pintscher -
http://about.me/lydia.pintscher
Product Manager for Wikidata
Wikimedia Deutschland e.V.
Tempelhofer Ufer 23-24
10963 Berlin
www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg
unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das
Finanzamt für Körperschaften I Berlin, Steuernummer 27/029/42207.
_______________________________________________
Wikidata mailing list
Wikidata(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata
_______________________________________________
Wiki-research-l mailing list
Wiki-research-l(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wiki-research-l
_______________________________________________
Wikidata mailing list
Wikidata(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata
_______________________________________________
Wikidata mailing list
Wikidata(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata
_______________________________________________
Wikidata mailing list
Wikidata(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata