Dear all,
Thanks, Romaine, for this detailed and careful analysis of the
situation. I think much of this is spot-on. I think one of the main
insights here is that we need more uniformity. Wikidata in many places
is still used like some exotic "structured" format for entering plain
texts, which make sense to human readers but prevent or confuse
automated usage. The key is to "see" collections of items rather than
single pages.
It seems Wikidata would need more stakeholder communities for specific
areas (say sports events) to oversee and guide the modeling of the items
in this kind. We need more WikiProjects.
Regarding the question whether solutions need to be technical or social,
I'd say both must go together. I also have often been disheartened by
the sheer effort that it would require to add even the most obvious
statements to a larger set of items. Geography is a good example: there
are so many nearby places that share the same geo-administrative history
(take a look at the country, P17, of Dresden, Q1731), yet it is
practically impossible to add this to any significant amount of the
thousands of Germany cities ... Here, like in many of the cases Romaine
has described, the technical limitations may smother necessary community
activity. (The specific case might also be an example of something where
an approach of "data sharing" is needed, i.e. a modeling paradigm that
simply allows us to say "this place has the same history of P17
statements as this other place"; but that's not the main topic of this
post).
New tools may also enable and encourage communities to grow that have
not formed in the past decade. One aspect here might be that it is
difficult for communities to appreciate the result of their efforts. For
example, it is very difficult to create a uniform appearance for a group
of pages, already since the order of statements (in a group of the same
property) is so hard to change, and also since the pages are already
very long. Even if one can achieve complete semantic uniformity, one
will not currently have much opportunity to "see" this success. There
are unsolved challenges here that cannot be compared with the relatively
simple and small data that one can find in a typical Wikipedia Infobox.
External developers and maybe even researchers could contribute here,
but they would also benefit form the input and concrete ideas from
WikiProjects (Romain's email already had quite a number of directly
implementable ideas in it ... this kind of constructive input is already
half of the solution).
Cheers,
Markus
On 31/10/2022 23:40, Romaine Wiki wrote:
Yesterday it was 10 years ago when Wikidata was
founded and two weeks
ago Wikidata reached the amount of 100 million items. This is a good
moment to see what we have (and don't have), to look a bit back, and
also some hope for the future.
The idea to describe this already started in September and since then I
have done various analysis to get a picture. This, however, will not be
a complete overview as there are too many factors involved, just a
general picture of what I came across.
(Spoiler: This e-mail gets more structure further below. :-p)
== Structured? ==
Wikidata, it is said it contains structured data. I think we need to be
more precise with it: it is how the data is stored that is structured.
And this structured data is _only_ present on an individual item. If we
zoom out a little bit, and view multiple items of a serie, among items
the data is often missing, fragmented, differently organised, and
sometimes even problematic. On a multi-item-level (serie-level) it
highly depends if a user has done all the work to synchronise the
various items all together or not.
*Example:* I came across a serie of items about a certain sports
tournament with an edition organised each year for 50 years on a row.
For P31 (instance of), on 5 items it was called an event, on 25 items it
was called a sporting event, on on 13 items a tournament, on some others
a competition, and a few without P31. To be clear, each edition had the
same setup, was for the same sport, everything the same. The articles on
Wikipedia are better structured!
This is just a simple serie of items. Zooming out another level, the
differences between series are huge, which makes the quality low.
How is a new item added? In the past ten years many items have been
added with bots/tools based on the articles on Wikipedia. (Yes, for I
ignore here other additions.) In future still many items will be created
when an article on Wikipedia has been created. In the worst case, the
user adds the sitelink and the items stays empty (practically useless!).
A little bit better, the user adds P31/P279 (instance of/subclass of)
(not useful, but it helps). A bit more better, also other statements are
added (an item becomes useful). Better when a user checks one/two other
items in a series. Much better when a user checks all items of the row
of subjects. And fantastic when a user checks all items in a series and
in other series.
Realistic for most new items? No, this is way too much effort. At the
same time, to get quality data, it is needed.
*Example:* About a month ago there were 13 000 items with a sitelink to
the Dutch Wikipedia without the basic statements P31/P279. This is just
one language version, we have hundreds of wikis!
After some time after a new article has been written, users use a
bot/tool to mass import new articles from Wikipedia to Wikidata with
zero/little statements. We should be happy that they do this work, but
these items are largely empty and do not contain useful/needed data.
Also many duplicates are created this way. We need to go to the source
and find a solution there, re-thinking the workflow, otherwise we keep
mopping with the tap open.
*Needed for the future:* a "new article to Wikidata wizard". I imagine
that when a user is ready with writing an article, he clicks on Publish
page. As soon as the page is saved the user gets a pop-up dialogue. The
user is first asked (in the dialogue) to search in Wikidata to see if
already an item exists about this subject. With a completely new subject
or empty item, the second step is that the dialogue suggests (based on
the published article) a few statements the user can click and confirm.
Most new articles are about subjects that are part of some sort of
series or about a subject with a default set of properties we expect to
be always present (like a building: country, located in the
administrative territorial entity and coordinates).
I think we can be more precise about what Wikidata contains: it contains
chaotic data in a structured way, which is often not structurally added
nor maintained.
To get more quality, we not only must have the data structured on items
and among items, but also the way how we think about working with the
data needs more structure. We currently work with individual items, and
without an integral perspective on the data: we have no overview.
== Wikidata gives no overview ==
I sometimes heard users say that Wikidata can provide an overview. That
is however not true. Wikidata does not give an overview! Wikidata can't
give an overview itself, but a tool can create an overview with the use
of data from Wikidata.
To get more quality on Wikidata, more overview is needed. Overview over
what is missing but should have been added on every item of a series.
Overview over what unexpected use of properties can be found in a series
of items. Tools that currently exist are especially good in detecting
what data has been added, but not what data is missing or is weird for
this type of item.
*Needed for the future:* a tool "get me more like this item", but I
prefer to call it a "smart tool". When looking at an item, I often find
myself wondering about what other items of this series has as
statements.If a series contains 50(+) items, I have to open every single
item to see if anything weird is going on or anything is missing in
these items. I wish I could press a button "get me more like this". The
tool then shows a full series of items with the same label (like only
the year changes) (but also takes into account the labels in multiple
languages at the same time) and or with the same description and or with
the same/similar properties. The tool gives suggestions what to include,
but it is also possible to indicate that the tool should ignore certain
things. In this way I can easily find a certain sports tournament with
50 editions the past 50 years. And then includes also those editions of
that tournament that have no article (on WP) in my native language (and
thus no label in my language), but have an edition in for example in the
Italian WP. The tools shows all the properties added, without having to
indicate myself which properties should be shown, and can show the
labels and descriptions in multiple languages.
== Labels, descriptions and aliases ==
If I have to describe one of the main things I do on Wikidata it would
be fixing language. The number one thing to fix are capitals -> lower
case. I click edit, change the capital of the label, change the capital
of the description (if it is only one), and often changing the capitals
from all the aliases, click save. This sounds not much work, but with
visiting 100 items, it becomes a lot of work. And this was just one
language, often I fix it for English and French too. Can't this be made
easier?
*Needed for the future:* a tool with what I can fix capitals in one
click. In 99,9% of the cases they are capitals that need to be fixed to
lower case. Especially ligatures take a lot of work. If someone works on
this, take into account the ligatures
<https://en.wikipedia.org/wiki/Ligature_(writing)> and for Dutch also IJ
-> ij.
*Needed for the future:* a Wikidata game that can easily find items
where capitals are used while it should be lower case.
With many subjects the labels and descriptions are all right or all
wrong if it comes to capitals. One group of subjects is more
challenging, but in number as in the combination lower case/capitals:
taxons. Many labels got imported from Wikipedia. In Dutch for example,
the local names should be lower case and the scientific names with a
capital. This is currently a big mess on thousands of items.
Another thing I have to fix frequently are dots in descriptions.
Apparently some users like to use a dot in there, while they shouldn't.
Finding the places where this took place is very hard...
*Needed for the future:* being able to run a query on the labels,
descriptions and aliases. Many errors and issues can be find in their
and need to get solved, but finding them is not easy. I recently came
across a series of items with a spelling error.
Did you know that there are more than 20 places in the world that are
called Amsterdam? How useful is then a description "building in
Amsterdam"? Yes, a large number of users find it too much work to add
the country of where a certain item is located.
*Needed for the future:* a tool/query with what I can quickly get an
overview of all the descriptions that doesn't contain a country.
*Needed for the future:* a Wikidata game that gives me descriptions
without country while they should have one.
We have arrived at useful labels and descriptions. A lot of work needs
to be done in that field. Many subjects do not have a unique label as
there are other subjects with the same name. To select the right item, a
description is needed to clarify the context of that item.
*Needed for the future:* a Wikidata game that can generate descriptions.
For many items the description can simply be <subject =P31> in
<location/administrative territorial entity =P131/P276>, <country> At
the same time this can be added in your local language as in English, so
everyone knows what the topic is about. (Bonus: there are still items
that do not contain a country, maybe something to be fixed right away?)
A Wikidata game can help to find items with missing labels/languages,
but it should be possible to simple query these.
On the other hand, I also have came across items with many wrong
descriptions, especially "Wikimedia category" and "Wikimedia
disambiguation page". Sometimes this can't be simply reverted causing a
lot of manual labour. On a recent occasion it took 50 minutes
<https://www.wikidata.org/w/index.php?title=Q89509298&action=history> to
get the page saved!
*Needed for the future:* a tool instantly removes in one item all the
labels of disambiguation pages, Wikimedia category or Wikimedia list
article.
Having at least a label in English is very welcome, otherwise there is
no clue what Q1234567 is about. There are bots who add missing labels,
including copying the page names from Commons. The sitelinks on Commons
are often Commons categories that are connected to items about that
individual subject. The bots adding the missing labels sadly also copy
the prefix Category:when entering the labels, which is often wrong.
Simple solution: Only add the Category: prefix if P31 has Wikimedia
category as statement.
I personally think that the biggest weakness of Wikidata are the missing
labels, and then in particular the missing labels in English. If an item
has no label at all, it is basically useless. If an item only has a
label on a local language (and not English), it only can be used in that
local language which is a minority of the world. At the same time, while
in many countries most people speak also English next to their local
language, in many other countries this is not the case and people don't
understand English. This is a matter of accessibility and therefore it
has priority.
*Needed for the future:* a program to get for (almost) all items a label
available in English + translations of this label in many local languages.
*Needed for the future:* the minimal requirement for batch uploads that
they contain at least a label in English.
*Needed for the future:* a tool that helps with translations. There are
many items with the same name. Currently we have to add a translation to
each single item. A tool would be handy to find all items with the same
name in English (like: Saint Servatius Church), and then being able to
add a translation only once which the tool add to all the items.
*Needed for the future:* a tool that can do transliteration.
Transliteration is a huge barrier for the usability of the data, as many
labels are only added in one script, while the user uses natively
another script. This especially involves names.
Especially smaller language communities have a hard time on Wikidata. A
small language community means that only a very limited part of the
(essential) items on Wikidata gets translated into their local language.
At the same time, if you work on adding statements to Wikidata (while
being a non-English speaker), you highly depend on translations being
available in your own language. If something has no label in your local
language -> it will not be found when searching with the local word ->
you can't add a statement or you likely add a wrong statement.
Without any statements an item is practically useless, so various users
are searching for items without any statements to add them. While doing
that, I recently came across a few items with only a sitelink to a
Wikipedia language version. This language is not available in Google
Translate or any other translation tool I could find, resulting in that
nothing could be done with these items.
== Statements ==
Every item on Wikidata should have have at least a statement instance of
or subclass of (P31/P279) (or both), because these two properties define
what the item is about. Without these properties, we are practically
blind. It is great to see some of you are working on getting all the
items to have these properties on them. (I recently completed that for
all items with a sitelink to nlwiki: 23 373x -> 0x.) More help is needed
for the many other items!
*Needed for the future:* a Wikidata game that brings up items without
P31/P279 and gives suggestion(s) to add.
While doing the project of adding P31/P279, I noticed that various users
still do not understand the difference between these two properties.
This means that on various items users have added the P31/P279 wrong. We
need to think on how we can find the items where that is the case and
fix those.
There is also a grey zone: a series of items for what it is not
precisely clear whether it should be. Perhaps a project who can take
care of those cases?
In addition to P31/P279, each theme of items has a fixed set of basic
statements that always should have been added. For example with taxon as
P31, also needed are scientific name (P225), taxon rank (P105), and
parent taxon (P171), For example with building as P31, also needed are
country (P17), administrative territorial entity (P131), and coordinates
(P625).
This seems pretty obvious, but a recent large data import still forgot
some basic statements, which still hasn't been fixed. The goal of adding
data is that the data can be used. By having some basic statements
missing, the quality becomes too low. I think such data imports should
not be allowed.
*Needed for the future:* a program/project to get for (almost) all items
the basic statements present.
We then probably should also have attention for the quality. For example
the administrative territorial entity (P131) should not be too generic.
A village or a building should get as P131 the smallest administrative
territorial entity as possible (in many countries the municipality).
With the various fixed sets of basic statements I estimate that it is
possible to cover at least 90% of all the items (taxa, geographic
features, people, structures, astronomical objects, publications, etc.).
The remaining ones are harder, often more specialistic. Those have often
for P31 "term" or "concept". To make those items more useable these
items need to get more statements that provide a better context. Then
properties like "aspect of" (P1269) and "characterized by" (P1552)
are
needed.
== Identifiers ==
About identifiers can't be said much: they do what do have to do. Even
within Wikidata they help a lot, as a symbol is shown when the same
identifier has been used on an other item en thus solving duplicates.
The focus where most identifiers seem to be related to are sports,
popular culture (music, movies, etc) and monument identifiers. In many
other fields no properties for identifiers have been created yet.
For a generic user, most work regarding identifiers is in finding out
how to find the specific identifier on which website so it can be added
to the item. I think there we should have more attention for so we can
make it easier for users to add them. Another thing is that for the
theme I am working on, it is not easy to see where identifiers are
missing but do likely exist.
*Needed for the future:* a tool that lists all potential items (in
general or of a set of items/query) where an identifier likely is missing.
If identifiers are added to items, an icon next to it often indicates
which statements are missing on that same item. For example, if I add a
monument identifier, it also indicates that for example a country (P17)
have been added. A great help to get items more complete. At the same
time identifiers are often forgotten. The other way round would be
welcome too: when an item gets for example building als P31, it should
also suggest to add a country (P17), an administrative territorial
entity (P131), coordinates (P625), and perhaps more.
== Other ==
Besides the ones already mentioned there are some tools/software/issues
that would make the work easier or need to be solved.
*Needed for the future:* a tool that looks up all coordinates nearby
certain coordinates. Like the Special:Nearby, but then any given location.
*Needed for the future:* better suggestions when adding statements. For
example, when I added bridges (those things to cross a river), I get
suggestions for properties related to astronomy. When an item has a
Wikipedia article as sitelink, it would be great if a statement
suggester would use the Wikipedia article to give suggestions. For
example, why do I have to indicate manually the country (again) if this
already has been indicated twice in the Wikipedia article?
Ten years ago Wikidata started. Those years past by quickly. We all
together have put so much work in it with a great result as outcome. But
we are not ready yet. For the next ten years I expect our main focus to
be improving the quality.
_______________________________________________
Wikidata mailing list -- wikidata(a)lists.wikimedia.org
Public archives at
https://lists.wikimedia.org/hyperkitty/list/wikidata@lists.wikimedia.org/me…
To unsubscribe send an email to wikidata-leave(a)lists.wikimedia.org
--
Prof. Dr. Markus Kroetzsch
Knowledge-Based Systems Group
Faculty of Computer Science
TU Dresden
+49 351 463 38486