Controlling quality for a database like this is tricky because
(i) not everybody wants the same things, (ii) people aren't willing to put in the elbow grease to solve quality problems that don't concern them (iii) people experience extreme psychological distress if you make them split hairs that don't matter
If you look at the ISO 3166 codes, for instance, these do change over time. There is a global and often a regional policy that "country borders should not change" but it is still going to happen. I know the ISO 3166-2 codes for 2nd level subdivisions are about to change in France because they have consolidated them recently. Worse yet, there are two letter codes that can stand in for country codes in various places, for instance there are some codes that can appear on passports that don't correspond to national governments.
I guess what it comes down to is any claim that a list is closed are situational and ought to be scoped to the opinion of someone or other.
On Wed, Mar 2, 2016 at 5:17 PM, Simon Razniewski razniewski@inf.unibz.it wrote:
Hi Michael,
Thanks for your comments! Some replies below:
I think an important discussion to have is that of identifying what
types of entities can or should be subject to the quality of completeness (or, for reasons to be explained, incompleteness).
Totally agree. As mentioned before, some properties are inherently fuzzy ("SignificantEvent", "Occupation", "AwardsReceived") and then completeness is hard to define. But others have a rather well-agreed definition (child as biological child, kings, bordering countries). Separating these classes will require effort.
As a very (very) contrived example, it may be the case in the future
that legally (because some fictitious court ruling) a "son" classification is reserved for only those who were conceived one year after their parents' marriage because of bizarre tax policy. How then is a prior classification of "compete" to be retroactively interpreted or reconciled?
I don't know the current policy of Wikidata regarding changes of concept definitions, but I guess one would need to do the same for all completeness statements that one would need to do for all regular facts that use the modified concept: Recheck whether they are still valid.
In practice and in the spirit of the Incompleteness Theorem it seems
like it could be more useful for the cool tool to be used specifically to identify entities which are incomplete, as opposed to complete. Incompleteness is a quality which can be known and gives us actionable steps for improving the quality of wikidata.
This is an interesting idea, we will see how to can add that to the demonstrator! Would be interesting to see whether more often people flag incomplete data, or directly complete it. A challenge in flagging data as incomplete is that knowledge of incompleteness in many cases implies knowing what is missing (I find examples of situations where one knows that data is missing, but doesn't know the data a bit contrived (except for functional properties)), so it would be interesting to see how such a feature would be liked.
Cheers,
Simon
On 2 March 2016 at 21:51, Michael Karpeles michael.karpeles@gmail.com wrote:
I think the cases Tom provides are significant (not in evaluating the usefulness of the cool tool, but in considering how the quality of completeness is used within Wikidata). A lax interpretation of completeness exposes an opportunity for subjective misclassifation of entities which is hard to detect and/or correct for (because subjectivity is by definition ambiguous) .
The COOL-WD tool itself seems useful and things like Godels Incompleteness Theorem should be considered as a cautionary guiding principle rather than a deterrent to progress ( https://michaelkarpeles.com/essays/philosophy/incompleteness-theorem). I think an important discussion to have is that of identifying what types of entities can or should be subject to the quality of completeness (or, for reasons to be explained, incompleteness).
Some concepts are indeed both discreet and finite and also are not subject to variation or stochasticism over time (read: not subject to mcarthy's frame problem). A game of checkers which is fully explorable, and unchanging in its rules, is an example of a type of entity whose problem space can be considered completely known.
But even in regarding an event which has already transpired, while one may argue the event itself is complete, this does not preclude new information coming to light (as Tom points out) which changes how we record this event. And if we really care that the event has already happened, that's probably better represented by its date field which is not subject to the same level of ambiguity.
As a very (very) contrived example, it may be the case in the future that legally (because some fictitious court ruling) a "son" classification is reserved for only those who were conceived one year after their parents' marriage because of bizarre tax policy. How then is a prior classification of "compete" to be retroactively interpreted or reconciled?
In practice and in the spirit of the Incompleteness Theorem it seems like it could be more useful for the cool tool to be used specifically to identify entities which are incomplete, as opposed to complete. Incompleteness is a quality which can be known and gives us actionable steps for improving the quality of wikidata.
Additionally it seems like it would be helpful (and maybe this is how it works, I haven't checked) if the quality of incompleteness included some citation so researchers don't spend hours trying to investigate a claim which may have been made accidentally. Tom,
how do we know whether anything is the truth? I would argue that for completeness statements, as discussed by James Heald above, we should use pretty much the same criteria we use for anything else - i.e. not truth, but whether the sources support that statement.
I.e. I don't see much difference in the question of truth for the statement "Barack is the father of Malia" or "Barack's children are Malia and Sasha and that's a complete list".
Cheers, Denny
On Wed, Mar 2, 2016 at 9:51 AM Michael Karpeles < michael.karpeles@gmail.com> wrote:
Godels incompleteness theorem, QED. On Mar 2, 2016 8:39 AM, "Tom Morris" tfmorris@gmail.com wrote:
I can see how one could measure completeness of article transcription in the 1911 edition of Encyclopedia Britannica, but I don't see how you can measure the completeness of a list of Obama's children, or anyone else's for that matter.
First, it's temporally sensitive, so it depends on when you ask or what point in time you want to know about. Second, speaking as an occasional family historian, the father may not know all his children or, even if he does, be willing to divulge the complete list.
This situation isn't unique to family composition, it applies to many, many aspects of the real world. How do you know something is "complete?" The fact that you've got everything in the latest Rembrandt catalog doesn't mean that you have a complete list of Rembrandt's paintings.
Tom
On Wed, Mar 2, 2016 at 11:27 AM, Fariz Darari fadirra@gmail.com wrote:
Hello Jane,
I did some look-up about the Encyclopedia Britannica on Wikisource (thanks for the pointer), and yes, some part is complete ( https://en.wikisource.org/wiki/1911_Encyclop%C3%A6dia_Britannica/Vol_1:1) and some part is yet to be completed ( https://en.wikisource.org/wiki/1911_Encyclop%C3%A6dia_Britannica/Vol_20:4 ).
With respect to that use case, COOL-WD could perform the following inference:
- Suppose that we are complete for all the volumes of the EB.
- Suppose that for each volume, we are also complete for all the
sections of the EB. 3. And last, for each section, we are complete for all the topics. Then, one conclusion by COOL-WD is that querying for all the topics of all the sections of all the volumes of the EB would give us the complete answers :)
As for the statistical inference (e.g., how far are we complete for the EB wrt. volumes, sections, and topics), this is indeed an interesting idea that could be featured on a next release of COOL-WD!
Regards, Fariz
On Wed, Mar 2, 2016 at 11:26 AM, Jane Darnell jane023@gmail.com wrote:
I had no problem reading your mail. Thinking it over, this would also be a way to track the connection of dictionaries in Wikisource to items in Wikidata. So for example, Wikisource has lots of imported articles from the Encyclopedia Britannica 1911, and it would be nice to track
- Completeness in Wikisource (how many articles are complete in
section "A"?) 2) Completeness of matchups in section "A" to articles in English Wikipedia (how many subjects of Wikisource EB 1911 "A" articles have items on Wikidata with a link to English Wikipedia)
Once you have all that, it would be interesting to know about the relative completeness of a) Places b) Events c) Male people d) Female people
etc.
On Wed, Mar 2, 2016 at 10:57 AM, Fariz Darari fadirra@gmail.com wrote:
> Hello Jane, > > thank you! Yes, that sounds like a suitable use case for COOL-WD! > > PS: Pardon the formatting of the announcement email, somehow the > linebreaks are vanished :( > I am now experimenting with another email client, hopefully the > linebreaks are there. > > Regards, > Fariz > > On Wed, Mar 2, 2016 at 10:38 AM, Jane Darnell jane023@gmail.com > wrote: > >> Wow this sounds great! I would love to use this for oeuvre catalogs >> of top painters! The latest Rembrandt catalog is complete on Wikidata, as >> well as a few other ones, but older ones are not yet complete. This could >> be a great tracking tool for WikiProjects. >> >> On Wed, Mar 2, 2016 at 10:17 AM, Darari Fariz < >> Fariz.Darari@stud-inf.unibz.it> wrote: >> >>> Hello Wikidata community! Wikidata is a great platform for >>> collecting information, and the high quality work of many authors yields >>> very reliable information. Still, a challenge for users of Wikidata is that >>> there is no way to see whether *all* data on a certain topic is in >>> Wikidata. For instance, it is easy to see that Malia and Sasha are children >>> of Obama, but there is no way to specify that these are all his children. >>> More generally, Wikidata stores many facts, but it stores no information >>> about which topic it contains all facts. Today we are happy to share with >>> you a prototype that allows to add and manage such completeness >>> information, and would be happy to get your feedback on how useful you >>> consider this tool, or where you see space for improvements. With our >>> prototype, called COOL-WD (Completeness Tool for Wikidata), one can: 1. See >>> completeness statements for Wikidata facts 2. Add, remove, aggregate and >>> filter completeness statements 3. See how completeness statements allow >>> conclusions about the completeness of SPARQL queries over Wikidata. COOL-WD >>> is available at http://cool-wd.inf.unibz.it/ and a 3-min demo >>> video can be found at http://cool-wd.inf.unibz.it/coolwd-hd.mp4 >>> It employs various libraries, most importantly GWT, Apache Jena, SQLite and >>> the Wikidata API. The formal background and description of the tool >>> including an indexing technique for completeness statements have been >>> accepted as a research paper at ICWE 2016 ( >>> http://icwe2016.inf.usi.ch/) available to download at: >>> http://bit.ly/1VOsRCH Below are some naive ideas of how >>> completeness could be useful to users: > Use Case 1: Rido is a geographer >>> who would like to contribute to Wikidata about the administrative divisions >>> of regions. He cares so much about data quality, especially data >>> completeness, and is collaborating with Simon, another geographer. However, >>> when completing data on Wikidata, there is currently no way to mark which >>> data is complete. Rido and Simon must make these notes about completeness >>> manually in, say, a Google Doc. Worse still, the effort from Rido and Simon >>> to complete data could not be appreciated by Wikidata users since to the >>> users’ eyes, there is no difference between complete data and incomplete >>> data on Wikidata. Demo: Wikidata is complete for all administrative >>> divisions of Saxony (http://cool-wd.inf.unibz.it/?p=Q1202) > Use >>> Case 2: Jen is a developer of a moviegoer application. She usually >>> integrates data between multiple sources including Wikidata. If some movies >>> on Wikidata have completeness statements, she might optimize her >>> application to not search in other data sources for those movies. Demo: So, >>> when her app is asking on COOL-WD at >>> http://cool-wd.inf.unibz.it/?p=query for cast and screenwriters >>> of the movie Before Sunset (http://cool-wd.inf.unibz.it/?p=Q652186): >>> SELECT * WHERE { wd:Q652186 wdt:P161 ?c . wd:Q652186 wdt:P58 ?s } Her app >>> gets not only query answers but also the completeness information of her >>> query. We are looking forward to your feedback! Best, Fariz, Simon, Rido, >>> and Werner Free University of Bozen-Bolzano, Italy >>> >>> _______________________________________________ >>> Wikidata mailing list >>> Wikidata@lists.wikimedia.org >>> https://lists.wikimedia.org/mailman/listinfo/wikidata >>> >>> >> >> _______________________________________________ >> Wikidata mailing list >> Wikidata@lists.wikimedia.org >> https://lists.wikimedia.org/mailman/listinfo/wikidata >> >> > > _______________________________________________ > Wikidata mailing list > Wikidata@lists.wikimedia.org > https://lists.wikimedia.org/mailman/listinfo/wikidata > >
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata
Wikidata mailing list Wikidata@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata