'On Fri, Jul 11, 2014 at 4:53 PM, Erik Moeller erik@wikimedia.org wrote:
On Thu, Jul 10, 2014 at 11:02 PM, John Mark Vandenberg jayvdb@gmail.com wrote:
Or .. sometimes the licensing and attribution information isnt correct
In the common case, Media Viewer provides more prominent and appropriate attribution and license information than the File: page. The author name, license, license URL, and source URL are all immediately accessible below the image, whereas on the File: page there are sometimes screenfuls of metadata between the image and this crucial information.
Im not suggesting that the layout of MV isnt great, but that isnt the issue I raised, and you are unfortunately very wrong and dumbing the licensing and attribution information down is fraught with problems that need to be carefully worked through,
Let's walk through a real example, of a page with over half a milion page views in the last two days.
https://en.wikipedia.org/wiki/Indonesian_presidential_election,_2014 http://stats.grok.se/en/latest/Indonesian_presidential_election,_2014
The photo of Prabowo in MV has a caption of 'Prabowo wapres' instead of 'Prabowo Subianto' which is (now) the caption and alt text on the enwiki article, and the the Jokowi photo in MV has a caption of 'Gubernur DKI Jokowi' despite having an {{information}} block on Commons with an English & Indonesian descriptions, albeit without perfect syntax, but that is par for the course and MV design needs to cater for this type of scenario.
Scrolling down on the Prabowo image in MV shows '== Licensing: ==' .. whoops wasnt wikitext supposed to be hidden? This is another syntax problem with the wikitext, but our goal should be to show broken wikitext to as many eyes as possible so that one person takes the initiative to try to fix it. The licensing shows 'CC-BY-SA 3.0', but the image is 'correctly' tagged as CC & GFDL. The Commons Slideshow gadget manages to correctly detect that this image has metadata that says it is dual licensed. Try https://commons.wikimedia.org/wiki/Help:Slideshow on https://commons.wikimedia.org/wiki/Category:Prabowo_Subianto . Why does the opt-in slideshow gadget have a better understanding of media metadata than the WMF opt-out media viewer?
Scrolling down on the Jokowi image in MV shows Indonesian text instead of English text, it isnt identified as Indonesian language despite very good syntax in the wikitext, and the 'License details' block on my screen shows the first two lines of the licensing template, and the top third of the third line of text which makes it unreadable, and a 'View more' sort-of-button which appears at the bottom right also overlapping with the second line of license text, obscuring that also. I can expand the box to show the rest of the license details, but .. eww .. this is out of beta really??
From an ideology perspective, these image pages have many issues which
needed to be edited, and the MV doesnt promote editing. It shows dubious, incorrect, or syntactically invalid metadata as if it is un-editable, instead of suggesting that the metadata needs editing. It doesnt highlight that one of these images is claimed to be a CC photo , yet it is a professional shot, is uploaded by someone with a red username. Our astute image reusers should be looking for an OTRS permission tag, which is missing.
The slidebar has a sequence which repeats the images: Probowo -> Jokowi -> Prabowo -> Jokowi -> Prabowo -> Jowoki and then whooping huge Indonesia flag colors.
I could easily double or triple what I have written above about just that one page, focusing on finer details of the MV UI that are poor design or inadequate implementation, but .. sigh .. you have people paid to do your designs.
And, there are also sorts of quite important parts of the licensing process which are ignored by MV. e.g. a image which is OTRS pending, now doesnt include a big scary warning one click away. https://en.wikipedia.org/wiki/Nick_Driver#mediaviewer/File:Nick_Driver_in_20...
Erik, do you know what percentage of image pages are not correctly parsed and presented by MV wrt licensing and attribution?
This is actually a pretty remarkable accomplishment given that this information comes from a huge number of different templates that vary across wikis. Media Viewer makes use of standardized CSS classes to extract metadata, and the team has actively worked with the community to broaden their use:
http://lists.wikimedia.org/pipermail/multimedia/2014-March/000135.html
Ultimately we'll want to use proper structured data for this, but these changes lay the groundwork, and there's already an API (used by Media Viewer but open to anyone) that exposes this information.
Where no license is detected, Media Viewer still falls back to a "View license" link. The more problematic cases are where actual errors occur and important information is not extracted, and there will certainly inevitably be some cases where this happens, but this can only be worked on over time. The expectation that an unbounded problem like this is completely solved prior to deployment of a feature is unreasonable -- it's similar to TemplateData, in that the positive feedback loop into Media Viewer should actually help encourage more and consistent use of machine-readable data.
sometimes you get resolutions which are silly (especially svgs at launch, but also slideshows on a file page include a very large license logo)
Can you give a specific, current example?
See above for the 'flag' example, but go to commons, click random file, open in MV, and scroll though all . Seriously Erik, this happens on almost any random use of the MV - you dont need me to tell you this do you?
it takes extra clicks to get to the full-size version,
It takes exactly one click using the "View original file" button.
You mean to say that after the launch, and after much anguish of the community telling the WMF that the extra click was a major pain point, this button was added. It is great that bugs are being fixed, but don't be surprised if you have a hard job advertising the fact you've fixed major failings of the a product at launch, and it would be humble of the WMF to use clear messaging when talking about post launch bug fixes rather than 'It does what you want' type messaging.