Wikidata:Requests for comment/Define lists on both "Wikimedia lists" and "Wikimedia categories"

An editor has requested the community to provide input on "Define lists on both "Wikimedia lists" and "Wikimedia categories"" via the Requests for comment (RFC) process. This is the discussion page regarding the issue.

If you have an opinion regarding this issue, feel free to comment below. Thank you!

THIS RFC IS CLOSED. Please do NOT vote nor add comments.

The following discussion is closed. Please do not modify it. Subsequent comments should be made in a new section. A summary of the conclusions reached follows.

Closing this RFC over a year after last comment. The following are the results of the RFC:

• items that are instances of Wikimedia category (Q4167836) can use is a list of (P360) to indicate the primary item (class) that is covered by that category, with qualifiers to allow automatic generation of a list. (See #Proposal 1 - Categories and #Update: one year on - One year further on there remain a large number of categories using is a list of (P360) in this way).
• If there is one wikidata item that links to both class pages and list pages on different wikis, that is ok; it should be labeled by the class with an appropriate subclass of (P279) statement and should not be indicated as instance of (P31) Wikimedia list article (Q13406463). However, it is also ok to have two separate items, one for the list and one for the class, which should then be linked with is a list of (P360). (See #Lists are classes ).

ArthurPSmith (talk) 15:32, 15 April 2016 (UTC)

---

Lists and categories are in effect the results of queries that are the result of a query. While we are waiting for the implementation of query functionality in Wikidata itself, queries are functional in WDQ. WDQ is the query tool used within Reasonator and is a tool in its own right as well.

At this moment we can show lists in "Wikimedia list articles". They do show in Reasonator.

The definition is "is a list of" (the instance or subclass) and the rest of the qualier are the pertinent conditions that have to be met. An example can be found here.

In this way, list and categories gain relevance. They can be used to compare the content of a list in a project with what we know in Wikidata. Thanks, GerardM (talk) 08:35, 22 February 2014 (UTC)

I think I see what you mean now.

We have lots of "Category:" items on Wikidata, created in response to the need to have sitelinks between WP Category pages but what does a Category mean to Wikidata? For example Category:Indian artists (Q8544697)?

GerardM's cool idea is to have each Category Qitem link to an equivalent query so Category:Indian artists (Q8544697) would correspond to a query for items with properties Instance of:human; Occupation:Artist; Country of Citizenship:India and if you asked reasonator about this item it would to put together such a query and automatically generate a list of items.

This means that we would therefore need to add these three statements to the Category:Indian artists (Q8544697) item and similarly for all other Category items so Reasonator would have the info it needs to put together the query.

And maybe something similar could be done with wikidata list items.

Was that it? Filceolaire (talk) 23:03, 25 February 2014 (UTC)

Imho it's better to have a class item, whom by definition have a set of instances. This set of instances may correspond to a query, like a list or a category usually do. Then we should link the categories or lists items to the class with the query. As we often have both a list or a category which should have the same set of instances, it seems a better solution to put just the query. Plus the subclass relationship looks a lot like the subcategory. Of course categories are less well defined and they sometime looks more like the informations about close items reasonator sometime shows. TomT0m (talk) 23:14, 25 February 2014 (UTC)

Tom, for many categories and lists there IS not class. An example is "Indian writers", "Painters from Pittsburg", "waffles from Brussels". GerardM (talk) 05:51, 26 February 2014 (UTC)

@GerardM: well there should be no intersect classes like these on wikidata. I'm not sure we don't have any.

Intersection classes are not a problem. They were a problem in Wikipedia as the category were too poor a system. In Wikidata, we can do better, and it's a feature we can define a class wrt. its properties imho. We can technically handle this and this allows to use this powerfulness to model things, and not to make a war of intersection categorist versus non intersection categorists. We can also see this as a way to tag queries with a name and use this query as class. Then if we have a property which as a specific domain, for example the swedish professors, we don't have to catch a headeach to know how we define the range or domain. Oh, the range is <person> as its a leaf of the class tree, but we will only accept a subset of this class for this property !! how will we solve this ?. We don't, we have a solution, we have a class, we have the corresponding query. Of course we do not have to make an item if we use some kind of class expression as in OWL, but I really don't think it's a good thing to make a rule of never create a class item when we can substitute it with a class expression. It's bad for reuse for example. To sort classes we have class classification, and I think it's a good practice to think in term of class classification to make things flexible and allow people to use the model they want to use and not use what they do not want to. TomT0m (talk) 11:29, 26 February 2014 (UTC)

@TomT0m: where a Category is a class the query is easy to define - instance of:'class'. Personally I think we should not do the 'automatic query' things for lists because lists have other problems, as discussed below. Wikipedia Category may (edited to add->not be well defined but Wikidata Categories don't have to follow that. WP Categories and WD Categories can be different (though similar) things and WD lists can be strictly defined by their statements even if WP Categories aren't. Filceolaire (talk) 06:25, 26 February 2014 (UTC)

The distinction class/nonclass is a headheac. Better not take that path, see my answer above. TomT0m (talk)

Proposal 1 - Categories

Add statements to each Wikidata Category item to define the query that can automatically generate a list. GerardM tells me the Reasonator is ready to go. See this. Filceolaire (talk) 06:25, 26 February 2014 (UTC)

Discussion

 Support Filceolaire (talk) 06:25, 26 February 2014 (UTC)

 Support Could be slightly more general than categories; any item with a "list of" property can be used for this. --Magnus Manske (talk) 00:43, 2 March 2014 (UTC)

Proposal 2 - Lists

Use 'instance of' or 'part of' or a new property to link items to List items to create hand compiled sort of classes. Reasonator can do a query on 'new property:list of Foo' to get the list. This preserves the distinction between hand compiled Lists and machine compiled Categories in a wikidata way. Filceolaire (talk) 06:25, 26 February 2014 (UTC)

Discussion

 Comment I'm not sure if this is a good idea but I thought I should spell it out as a proposal for discussion and see what others had to say. Filceolaire (talk) 06:25, 26 February 2014 (UTC)

 Oppose We are already doing that - Reasonator will show these lists (check "from related items" here). But, this is not what automatic lists are about. Those are manual lists. Which we should have too. --Magnus Manske (talk) 00:47, 2 March 2014 (UTC)

 Oppose. Thinking about it I'm against having any 'List of' items in Wikidata. I think the wikipedia 'List of' pages should be sitelinked to wikidata Class items i.e. items with the 'subclass of' property or other equivalent properties (such as 'Parent Taxon'). Filceolaire (talk) 10:03, 6 March 2014 (UTC)

interwiki conflicts

I want to make it clear that we often have interwiki-conflicts in these items. Here is an example. The English article was only a "list of governors" at that time, while the Swedish article described the "county". (This specific case has been solved both on enwp and dewp since then.) I think we should be carefull when we use Wikidata for interwiki in these articles. If one article is a list and the other isn't, then they should not share item. Many WP-users will not be happy with this opinion. -- Lavallen (talk) 11:23, 22 February 2014 (UTC)

It is OK when they do not like it. It is very much beside the point. When something is a "Wikimedia list article" is means that it is valid to define the list. GerardM (talk) 12:05, 22 February 2014 (UTC)

Hi GerardM, a question on the tool : If there exists a class indian film director, which some indian is an instance of, can reasonator (or anything else) be used to link this class to the query defined by the is a list of qualifiers ? How will this interract with the future query namespace ? It does not seem clear how queries will be defined and this way of doing might or might not match. But the list/instance link seems pretty clear, I'm a bit surprised as when instance of (P31) is becoming more and more important tool we introduce we do not follow the path of basing more things on it. TomT0m (talk) 15:05, 22 February 2014 (UTC)

There is no need for such a class. It is just a list / category that uses a combination for occupation and nationality.. IMHO having a class like that is silly. As to the future queries, they do not exist and I do not speculate. What I do is consult with Lydia and ask her opinion. I did for this as well. As far as I am concerned, the fact that we do include the "instance of" or "subclass" has everything to do with how you make this work.. A query would work almost equally well without them.. Artificial persons would show up.. Thanks, GerardM (talk) 22:45, 22 February 2014 (UTC)

Look at your answer to Joe below, the King of foo class is quite the same. There is plenty of cases where the distinction between a class and a query is quite blurry and a tough decision to make. A standard modeling language like OWL allows to define classes not only through the subclass of relation but also by a query on these properties. It is useful also for defining the domain and ranges of properties : the range human is quite unprecise for example if we want to model that only a french citizen can be a candidate of french presidential elections. I'm quite dubious that saying there is classes and there is queries, the two do not mixes this would be silly will help on any way. It just will be a pain to decide sometimes and a conflict source. TomT0m (talk) 23:25, 22 February 2014 (UTC)

Tom, you conflate to things. This is where we discuss problems with interwiki links. When incompatible articles are combined in one item, they need to be split. A list is obviously different from what it is a list of. Queries are defined on lists. Articles on a subject may include lists.

Conflicts are normal. It is a matter of perspective. However, when you are talking about academic tools like OWL, they will insist on normalising the data. Even academically this should be obvious. Thanks, GerardM (talk) 10:08, 23 February 2014 (UTC)

I don't agree. OWL is exactly the kind of technology who can help avoid conflicts by allowing to use different viewpoints on how to represent the datas while allowing to work on them as easily. This example on silly classes is a good example : they are silly for you, not for me, OWL would allow us to work with this without conflict and without imformation loss. It's a wày to smooth conflict. TomT0m (talk) 12:40, 23 February 2014 (UTC)

Different subjects have different levels of details in the set of articles. Some projects merge articles in some subjects, and split them in other. A few titles compared with the large set of items, have both articles about the title itself and an article with a list of all persons having that title. I therefor think we always need to have both kinds of items in the whole set of different titles, no matter if there are any articles or not. -- Lavallen (talk) 19:24, 23 February 2014 (UTC)

What is discussed in this section is the disparity of interwiki links where some are lists and others are not. Both of them are in the same item. Please explain how OWL will help solve this issue. Thanks, GerardM (talk) 18:57, 23 February 2014 (UTC)

@Lavallen, GerardM: If there is a list article and an article about the class, everything is clear. Unfortunately some articles about the class are judged list article or even worse disambig articles in other cases, which makes a big mess to clean up. Formally a list article lists a subset of the instances of a class, but sometimes there is text annotations to the items ad an introduction, and this makes a bit fuzzy the limit, especially when there is no corresponding article about the class. Conversely a typical class article may include a sample list of instances. To restore the interwikis when there is two item how about a template generating a link to reasonator using is a list of to find the relevant items, insested into the articles who have a lit <-> associated class ? Reasonator can handle this (that and generate a list of instances). I guess the difference beetween a list of instances class/named query and a list articles is that the items shown in the list article can be chosen, ordered and commented. TomT0m (talk) 20:43, 23 February 2014 (UTC)

As GerardM says in the header. A pure list and a category is something closly related. On svwp such lists are often deleted, and sometimes redirected to the category. I would like to see how querys will look like in Wikidata, not in an external resource like Reasonator before I tell exactly how we should act in this matter. -- Lavallen (talk) 20:09, 24 February 2014 (UTC)

Lists are classes

Summary agreed after skyping with GerardM:

Classes are essential to Wikidata. Classes will have the 'subclass of' property and may have the 'instance of' property (for special types of class Ford Mondeo:Subclass of:car; Instance of:car model). These can have links to appropriate WP pages which describe the Class. Some of these may list every instance of the class. Some of the WP pages may be called 'List of...'. That doesn't change the fact that the wikidata item is for a class and it should never have the property instance of:Wikimedia list.

In a few cases the WP page will get a long discussion of the class and they will decide to break out the List of page as a separate WP page. This can lead to there being 2 wikidata Qitems, one of which has properties which describe the Class and one of which has properties which describe the list, including instance of:Wikimedia list. Down the road a bit it would be a nice feature if 'List' items include a link to a query which would create a list based on Wikidata items that should be on that list.

OK?Filceolaire (talk) 13:01, 25 February 2014 (UTC)

Hmm Collapse top/Collapse bottom didn't work the way I thought. Anyone know how to fix that? Filceolaire (talk) 13:04, 25 February 2014 (UTC)

fixed Filceolaire (talk) 23:03, 25 February 2014 (UTC)

Not really OK, the query should be associated to the class. If we can associate a query to a list item, and that list item is linked to a class item, then the query defines the instances of the class. So it is the definition of the class itself or a part of it, hence it should be linked directly to the class. Apart from that question I think there is no real problem to solve at that point into this RfC as long as the Wikibase query are not implemented, to agree with Izno. That's all. TomT0m (talk) 18:33, 25 February 2014 (UTC)

I absolutely agree with Gerard. What you (and others) are failing to distinguish is that a class has a certain set of properties which may be shared by its instances, while a list enumerates the members of that class (or multiple classes).

I would have no objection to distinguishing between a plural/list page being used as the list page, while a singular page-name is used to represent the "unlinked" but more proper subclass claim.

Example: "king of Sweden" gets no wikilinks but does carry the subclass of "king" claim. A page "list of Swedish kings" links to the former via the "is a list of" property. --Izno (talk) 23:32, 24 February 2014 (UTC)

I don't disagree that a class has properties. I just fail to see why a class can't have sitelinks to a WP page which has a one sentence intro and a list of the members of that class, and there is no other WP page dealing with that class.

I promise you I have thought about this and it was working with actual wikidata Qitems, assigning properties, that led me to this conclusion. Filceolaire (talk) 08:48, 25 February 2014 (UTC)

If we were to look at this from a computing perspective, a subclass should know nothing about its instances. If we're classifying lists (which is what they really are!!!) as the subclasses, then we are implicitly doing so...

But that aside, the reason we should not is because "monarchs of Sweden" is not the head of Sweden. "monarch of Sweden" is. Once you change it from the plural to the singular, no longer are you referring to the list or plurality of monarchs, you are referring to a single one, or the class of those persons. External users will not expect the related content to be about all of the monarchs in an item titled "monarch of Sweden".

That aside, there is also the "list of" claim combined with P31 "list". You have not shown that it is helpful, and I would certain treat it as harmful, to use such claim on an item in the subclass tree. That is another thing that external users will not expect. Fundamentally, it comes down to how we classify our items, and we should do so in an easily-expected fashion. There should be a list item and a class item. I don't really care where the links get held (if they get held in the list item, fine by me!), but given that some topics have both a list and a class item and some topics only have the class item, this is the most deterministic option available to us. There is also the subject of categories, which should be linking not to list items but to the class items! --Izno (talk) 00:02, 26 February 2014 (UTC)

@Izno: Where are you reading? I see Monarch of Sweden (Q1268572) as the item describing the title "king/queen of Sweden" and list of Swedish monarchs (Q182162) describing the list. -- Lavallen (talk) 11:38, 25 February 2014 (UTC)

a monarchy is an organisation, a monarch is the one who is at the head of this organisation, so I don't see the monarchy as the set of all monarchs. It seems to be a wrong way to define things in this case. TomT0m (talk) 21:52, 25 February 2014 (UTC)

@Lavallen: It was just a convenient example off the top of my head which doesn't actually make my point. A better one might be any random list of fictional characters, with P31 list, which are being re-classified as P279 fictional character and having the P31 claim removed. --Izno (talk) 23:53, 25 February 2014 (UTC)

Speaking of organisations, there is also Royal Court of Sweden (Q6338017)... -- Lavallen (talk) 06:36, 26 February 2014 (UTC)

• GerardM's summary is pretty much what I've been doing. I don't see much advantage to, for example, having one item called "List of Seinfeld episodes" that has links to all the Wikipedia articles and a second item called "character from Seinfeld" that the characters' items all link to. I only keep the list and class items separate if a Wikipedia does and we need the inter-language links. --Arctic.gnome (talk) 03:32, 27 February 2014 (UTC)

Even a disambiguation page can, in some cases, be a class

I better get it out off my chest now. I believe there are cases where a Disambiguation page can be a class.

Most disambiguation (DB) pages list items with the same or similar names which are otherwise unrelated but some list items which have the same name because they are related. Typical cases (numbered for your commenting convenience):

1. 'Battle of Bar' disambiguation linking to 'First battle of Bar' and 'Second battle of Bar'. Often you find other WPs have one page discussing both battles rather than two separate pages and a DB.
2. Surname pages. Some DB pages have a list entirely composed of people with the same surname (maybe with a couple of places named after people with this surname) but there is no separate page discussing the origin/popularity of that surname nor ever will there ever be because the obvious place for this info is as a lead paragraph on the DB page. Why can't the family name (P734) property point to these pages? Once I labelled these as instances of Q11651459 but the deletionists have got rid of that. Oh well.

Filceolaire (talk) 09:07, 25 February 2014 (UTC)

As far as I am concerned, it is irrelevant what Wikipedia says. There are lists that are unstructured, there are categories that are unstructured and why not disambiguation pages as well. The point is that when there is structure it is explicit. When an item is defined as "is a list of" it is a list of something definable. This is how it is made explicit.

When something is made explicit as a list or a subclass, it has implications for the labels used. What is linked to an item is a secondary discussion. GerardM (talk) 10:03, 25 February 2014 (UTC)

Not to me. Wikidata needs a Qitem for each Class to act as the range of various properties. These Qitems can have sitelinks. If the only WP pages related to that class have a one line intro and a list of the class members then can we include a sitelink to that WP page on the class Qitem? If the answer is yes then most lists disappear from wikidata because they won't have a Qitem separate from the class Qitem. This shouldn't be a problem because wikidata doesn't have a semantic need (that I can see) for list Qitems. Filceolaire (talk) 10:51, 25 February 2014 (UTC)

No, Wikidata does not have that need. The proposition to generate lists from a set of qualifier in Wikidata looks damn well to me like a Query definition language and this query language is a subset of a class expression as defined in OWL. That said I'm not opposed to several ways of accessing the datas, but we should stop at some point and ask ourself if we are building something that is readable to an outsider ... I agree with you things are getting a little bit confuse at that point. TomT0m (talk) 11:02, 25 February 2014 (UTC)

You do not understand the proposal. The proposal is that when an item is defined as a list, we can define the query that defines the values of that list. Having lists is optional. I do not really care for lists. GerardM (talk) 06:18, 27 February 2014 (UTC)

I do understant that. Except a query is something more valuable you seem to think, and should not be attached to the list item. It should be the definition of the class of items this query defines. For example if we got

<Jerry> who is a character of the <Seinfeld> american TV series, we may have :

• a claim
  ⟨ Jerry ⟩ character of Search ⟨ Seinfeld ⟩
   ;
  ⟨ George ⟩ character of Search ⟨ Seinfeld ⟩
   ; ...
• a list <list of Seinfeld characters> with wikitext
  * Jerry
  * George
  to which you would attach the query : All Xs such that X Character of : <Seinfeld>.
• a class <Seinfeld character> ;
  that could be use for example to define a class of events <Seinfeld Costume party> from whom participants are supposed to be disguised in an instance of this class.

You say

⟨ list of Seinfeld characters ⟩ is a list of (P360) ⟨ All Xs such that X Character of : <Seinfeld> ⟩

. I say "the query defines the class". So instead of attaching the query to the list item, I say

⟨ list of Seinfeld characters ⟩ is a list of (P360) ⟨ Seinfeld Character ⟩

and

⟨ Seinfeld Character ⟩ is defined by Search ⟨ All Xs such that X Character of : <Seinfeld> ⟩

Then to be complete, we can add the claim

⟨ Jerry ⟩ instance of (P31) ⟨ Seinfeld Character ⟩

, but this may be inferred (for example if we add

⟨ Seinfeld Character ⟩ subclass of (P279) ⟨ Fictional Character ⟩

).

Funny fact

I just found in my watchlist the deletion of list of Doctor Who episodes (2005–present) (Q15728866), merged apparently by Epìdosis with (RfD: Merged with épisode de Doctor Who (list of Doctor Who episodes (2005–present) (Q387513))). Funny to see it's appening exactly when this discussion is occuring, but it seems that one of the page was a list of episodes. It seems that things have their own life independantly from decision making in Wikidata :). Don't really know what to do with this fact alone though  – The preceding unsigned comment was added by TomT0m (talk • contribs).

Lists and categories as queries

For me there is no much difference between categories, lists and queries, but of course some differences can be agreed on to know the scope of each concept. IMHO:

• Lists are pre-defined queries, the results don't need to be the same as manual Wikipedia lists
• Categories are "zoomable" queries, where you have hints in which direction to "zoom in" (sub-categories) and therefore need some sort of hinted structure
• Some lists have an equivalent category, maybe it makes sense to link them with a "same as"

So in general I find the proposed method appropriate, and I would even go further to extend it to categories. It is going to be hard to reproduce a categorization tree structure in Wikidata because category trees are very flexible, language dependent, and change over time, however the software could pool the sub-categories proposed by each Wikipedia and suggest some where there is some sort of agreement. And if you want to be really brave, what do you think about renaming the property "is a list of" to "is a query of"? (Maybe after consulting the WD development team, they might have other ideas for queries).

If @Magnus Manske: has time, I would like to suggest a further development to the WDQ tool, that could work as follows:

1. Once the list has been generated, the user can enter a category of a language wikipedia.
2. Both lists are compared, and only the differences shown
3. The user can click a button add to Wikidata items the missing statements that would make the item appear on the list

That way we could keep track easily of missing items and add add the needed properties to all of them.--Micru (talk) 13:07, 3 March 2014 (UTC)

Actually I'm seeing that the Article List Generator is able to generate article lists from categories, but there is no way of knowing which one of those articles are missing certain wikidata properties.--Micru (talk) 19:32, 4 March 2014 (UTC)

Update: one year on

One year on, and the issue of whether or not to use is a list of (P360) to describe categories is again under discussion, this time at Property_talk:P360#P360_and_categories, following on from discussion at Wikidata:Project_chat#.22is_a_list_of.22_on_categories.

Since 2014, User:GerardM has added inclusion criterion descriptions using is a list of (P360) to the items for about 2000 categories, which play well with Reasonator, as predicted. (For example the Reasonator entry for Category:Photographers (Q6473572).

However there has been some challenge, since the label for P360 still reads "is a list of", and the property constraints still include "not to be used on categories". One question raised is that what has or has not been put in a particular category might be quite complex, so a first stab (or even a second stab) at a P360 may be introducing a statement onto Wikidata that is not in fact completely true. One point of view holds that even if not perfect, such statements can give a good and useful enough rationale to describe why most of the contents may have been placed in the category; that an automatically generating exceptions list can be automatically generated, which by highlighting the edge cases may provide exactly what is needed to allow the P360 to be refined; or to exclude the exception list from any data processing based on the category, which an editor may consider good enough; or alternaively to use to identify items that maybe should have missing statements added. On the other hand, an alternative point of view holds that statements anything less than strictly true risks misleading and dangerous consequences, and so should have no place on Wikidata.

Of the other approaches considered above, the idea of associating each category with a class item, so the members of the category would just be instances of that class, has not been persued. Although it would lead to a particular straightforward corresponding query, it is weighed against by at least two fundamental considerations. Firstly the multiplication of items that would be required, to associate with all the myriad entries in the (often ridiculed) multitude of intersection categories that exist. And secondly because of a strong site ethic away from adding a separate instance of (P31) for each class an item is a member of, but rather in general to instead try to have only a single {{P|31]}, expressing the item's most fundamental nature, with additional aspects specified in properties. In such a system the items related to a particular class can no longer be found just by looking for instance of (P31) $1, but instead need a query of the for P31 -> A with Pxxx-> $1 , with some overarching class for A (such as human (Q5)), plus some particular specific property Pxxx. Which is in fact just what is coded on the category item by P360. Jheald (talk) 02:23, 3 March 2015 (UTC)

The premise that you can model categories in Wikidata is wrong. The arbitrary parts of a category are Wikipedia specific and should therefore not be included in Wikidata. What you can do is indicate for many types of categories are that they are a list of something. That something can be defined. Whatever fulfils that definition should be in the category (from a Wikidata perspective). In the end it is for Wikipedias to do whatever. Thanks, GerardM (talk) 10:23, 3 March 2015 (UTC)

---

The above discussion is preserved as an archive. Please do not modify it. Subsequent comments should be made in a new section.