New subject: Testing Wikidata arbitrary access (was: Re: Thinking about Structured Data on Commons ?)

15 Aug 2014

So now Wikimania has been and gone, can we think about where we're at 
with the Structured Data initative for Commons ?

In particular Liam, I think you said after one of the sessions you were 
"all over this" -- it would be good to know your thoughts.

I do think, as the GW toolset community, we ought to have a lot we 
should be able to offer here, because essentially we are doing big 
uploads from data which is *already* structured, so

(i)    we've got at least some experience already with working with data 
that is at least in some form structured
(ii)   we may know and be able to flag some awkward edge cases
(iii)  we would like to accompany uploads with data that can be "born 
structured", rather than converted later
(iv)   in any case we're uploading a lot of images, which somebody is 
going to have to convert to structured
(v)    we may have seen (or even written) some of the gnarlier templates 
on Commons, that migration will have to cope with.

It's not clear (at least not yet to me) how the Multimedia and Wikidata 
teams may best want to be communicated with, but I'm including Keegan 
(WMF) in cc:, who I think is the staffer with assigned community liaison 
responsibility.

The biggest message to me from looking through some of the documents 
after the meetings is just how much of the information is going to be 
stored as part of central main Wikidata.

Essentially, if we upload an image of an object, then it is expected 
that an 'item' (ie a Q-number) for that object will be added to 
Wikidata, which will contain all the metadata that describes the object 
rather than just the image.

The Wikidata community is already developing a very strong ontology to 
describe such objects -- key resources are

https://www.wikidata.org/wiki/Wikidata:WikiProject_Visual_arts/Item_structu…
https://www.wikidata.org/wiki/Wikidata:WikiProject_Books

where there are active and friendly communities involved in refining them.

We can get involved and help the process right now, by trying to 
identify and fill and gaps in these ontologies, and by being 
enthusiastic early adopters -- there is no reason we should not be 
getting involved right now, filling in appropriate metadata on Wikidata 
right now each time we upload an image to Commons -- real-world testing 
the current ontologies to see what creaks.

Data specific to the image itself (rather than what it shows) will be 
stored in a separate Commons Wikibase.

This will include such things as the file name, a file description, 
photographer, wikicontributor name, precise geographical location etc.

Commons Wikibase is also likely to contain a tag-like "topic list" -- a 
list of all the Wikidata Q-numbers that apply to the image.  These I 
think will be gathered by climbing up the Wikidata tree from any 
specified Subject identified for the image -- so a view of Westminster 
Abbey might get topics such as "Westminster; London; England; Cathedral; 
religious building" etc;  and games will be invented to encourage people 
to identify more such topics for the best images.

This should allow WM to introduced a proper combinatorial search engine 
based on tags for Commons; and many of the most egregious Commons 
intersection categories will wither on the vine.  (There is debate as to 
whether Commons will end up needing *any* category pages, but I suspect 
it will, because they are just so convenient to use as places for 
jotting down facts -- on the other hand, it is possible one might be 
forced to create an associated Commons article/gallery for that).

It would be nice (IMO) if there could be an interface to the topic list 
through the wikisource code for the filepage -- I think this would be 
well-received by the community, allow easy adaptation of existing bots, 
etc.  But this may be resisted as being too fragile a point of failure, 
as it would mean that people making hand-edits would have to know (and 
get right) the meaningless number-strings of individual Q-numbers.

Finally some very specific text data -- such as the EXIF data describing 
shutter-speed etc, is likely to continue to live on the file description 
page; because it's probably not something that people are primarily 
going to want to search, and it may be a bit unpredictable.

Part of the immediate effort in the next few weeks is going to be to 
produce clearer ideas about what information is going to live where, and 
in particular what information is going to live on the Commons Wikibase, 
and how it will be structured.

The good news is that much of the most complicated information will be 
stored on WikiData, so can be as detailed as we like (and can be 
accessed live now).

On the other hand, the design for Commons Wikibase will initially aim to 
be as simple as possible, with the aim to evolve it as experience is 
gained, to migrate the edge cases later.

The file description page (or something not entirely unlike it) will 
continue to exist as a view bringing together all the data.

Current templates will be re-written to draw information from Wikidata. 
  However, this won't yet be possible until the Wikidata team has 
implemented the "Arbitrary Access" feature -- the ability for a wikipage 
to access the properties of an arbitrary Wikidata Q-number.  What's 
causing the hold-up is that if the properties of the Q-number item are 
edited, then all the pages that access that Q-number need to be marked 
as dirty and regenerated.  That's easy if you only have one page that 
can access the Q-number, but hard if arbitrary pages can access it, 
through a chain of properties.

(eg: the file page for a painting Q12345 may use property Pnnn to its 
creator Q4567 who has property Pxxx, a date of birth.  If the date of 
birth gets made more precise, the system has to recurse back to indicate 
that all the file pages showing pictures of that creator's work need to 
be regererated.  This is tough, but file-page templates won't be able to 
draw on Wikidata information until it is in place).

It is progressively hoped to simplify the myriad of different templates 
used on the file pages as quickly as possible, to standardise them to 
draw from the structured data stores.

Templates to display summary information about collection objects, which 
will draw from Wikidata, may well be standardised so they can easily be 
used on Wikipedias and other wikis -- or, to put that the other way 
round, since Wikipedias and other wikis will also be developing 
standardised templates to display summary object information, it should 
well be possible to use the same code twice.

However, it would be good to get involved in the development of these 
templates, to make sure they accurately reflect the information we 
currently like to show in Commons.

(There may be some important details to get right -- for example the 
Wikidata data-type for dates currently comprises a 'best' value, and an 
optional numeric range (which is great for sorting).  But if the 
catalogue source data says eg "mid 17th century to early 18th century", 
do we want to make sure that precise string is still stored?  And should 
it still be possible to make it visible?  This needs close engagement; 
but probably principally with the community-based development effort in 
the Wikidata community groups.

Already very standardised are the present Commons creator templates and 
Commons institution templates.  These are likely to be an early quick win.

Looking down a typical present-day filepage, that means that it is the 
Source/Photographer information in the present "Artist" template, which 
is currently free-form and often a freely composed pull together of 
multiple different sources of metadata, that is likely to be going to 
need the most work to unpick.

This is also the field most commonly used for the credit link-back 
templates to the originating GLAM institutions, which are obviously a 
key consideration for our GLAM partners.

These templates may currently often be very institution-specific, and 
may do quite complex stuff -- eg the present version of the British Library
https://commons.wikimedia.org/wiki/Template:British_Library_image
as used at eg
https://commons.wikimedia.org/wiki/File:Cuthbert_discovers_piece_of_timber_…
contains link-backs to a number of catalogues, each with their own 
corresponding text; and as well as linking back to the information about 
the underlying object (which is likely to be stored on WikiData), it 
will also likely contain a link-back to the source of the original file 
(in this case the specific file at BL images online), which being 
information specific to the file is likely to be living on the Commons 
Wikibase.

The Source/Photographer field as a whole is (I think) likely to be one 
of the last on the file page to be assimilated, because it can be so 
sui-generis, and so the present rats nest of templates may continue to 
be acceptable for some time -- though even they are likely to need 
modification, as eg Photographer information moves to the Commons Wikibase.

That said, each institution is only going to need to manage its own 
template.

But it probably would make sense to start an effort to think
* what  is  the structured data that typically lives in these templates?
* and  is  there some standardisation we could start to get into the 
box, even now

Apart from anything else, something readily customisable might be much 
easier for new institutions to adapt and adopt.

For the migration project as a whole, an audit of all the source 
templates of this sort would be useful.  That is something the MM/WD 
project team could perhaps usefully encourage the community to undertake 
for them.

I have to admit there are lines I am not sure about, as to what gets a 
Wikidata entry and what does not.

For example, when does a photograph deserve its own entry?

Perhaps a bright line is that an image of a photograph one took oneself 
doesn't get an entry on Wikidata, but a photograph by Man Ray perhaps does.

What about a photograph by a photographer by more intermediate 
notability?  Or instead, perhaps an engraving from a book of 19th 
century engravings?

It makes sense to create an identifier for the book on Wikidata; and 
also the place depicted.  This is often almost enough to identify the 
particular image, but really one would want to store the page number, 
and perhaps the scan number as well. (Since one might well have either 
one or the other or both).  It would probably be good to store some 
identifier for the set of scans as well -- this too probably doesn't 
belong on Wikidata, (although one might identify it as set number 
<identifer> from eg the Mechanical Curator collection, which itself 
probably then *would* get a Wikidata identifier).

So the Commons wikibase probably needs to be able to identify images as 
having a sequence in a particular set, and that set as perhaps having an 
identifer that links it to a collection which has a particular  Q-number 
on Wikidata.

This is the kind of thinking we will particularly need to be doing over 
the next few weeks -- what is the metadata that will *not* be stored on 
Wikidata, so will *need* to be stored on the Commons Wikibase if it is 
to continue to be accessible?  That is something that we as the 
community need to evolve, thinking of all the use cases we can.

I am sure that there was something else I meant to say, but this email 
seems long enough already.

There's a scratchpad of some bookmarks I started keeping on a subpage of 
my userpage at Wikidata that people are welcome to,

https://www.wikidata.org/wiki/User:Jheald/bookmarks

This gives a nutshell of where some different fields might be stored

https://docs.google.com/presentation/d/1x-vOUr-zveLzoIP6uJC1Sz95xwuTFNwaBqJ…

This etherpad is good, esp lines immediately after 140, and "What new 
fields should be created to complement the old fields?" at 156 (actually 
in the context of Upload Wizard, but it gives some ideas)

http://etherpad.wikimedia.org/p/multimedia-wikidata-catchup

There's a spreadsheet showing some of the fields they're thinking about

https://docs.google.com/spreadsheets/d/1rk05EcLZpJaqOh5wymK6teIQufH9t0xn6oD…

-- though I think quite a lot of what's down as living on WikiData 
should really be Commons WikiBase --

and also a suggestion based on some simple use-cases:

https://docs.google.com/document/d/1C7UTB1kbaf_EisF3LmhpIQkb_ifSkB0rD8IGI9a…

though I think we would probably see that as *too* simple, even for a 
first build, because for many of our applications

     sequence-number in set
&   set-identifer in collection

are probably essential quantities to have (as they probably are for the 
WikiSource collection too).

Finally, this is an etherpad from the Hackathon just been, which has a 
lot of useful links at the end.
http://etherpad.wikimedia.org/p/structured-data-discussion-7-august-2014

Hope this initial brain dump is of at least some use, to make it worth 
its length,

All best,

    James Heald.   (User:Jheald).