As a basic rule for "which external identifiers are worth covering", I would begin with any  national identifiers we have for people (politicians, artists, writers, theologians, scientists, etc), then national identifiers for organizations (government-related, GNP-related businesses, nonprofits, educational institutions, etc), then national identifiers for places (census-defined population centers, battle-scenes, etc)

In my opnion, the question should not be "which identifier has the most coverage" but "which items have the most identifiers"


On Thu, Sep 7, 2017 at 9:26 PM, Andrew Gray <andrew@generalist.org.uk> wrote:
Hi Marco,

I guess this depends what you mean by "exhaustive". Exhaustive in that
every Wikidata item has ID X, or exhaustive in that we have every
instance of ID X in Wikidata?

The first is probably not going to happen, as the vast majority of
external identifiers have a defined scope for what they identify. Some
are pretty broad - VIAF is essentially "everyone who exists in a
library catalogue as an author or subject" - but still have a limit.
We're never really going to reach a situation where there is a single
identifier type that covers everyone, unless we're linking across to
another Wikidata-type comprehensive knowledgebase, and even then we'd
need to ensure we're in a position where they already cover everything
in Wikidata.

The second can (and has) been done - the largest one I know of offhand
for people is the Oxford DNB (60k items) but for non-people we have
complete coverage of eg Swedish district codes, P1841 (160k items).
It's a bit of a slog to get these completed and then maintained, since
the last 5-10% tend to be more challenging complicated cases, but one
or two determined people can make it happen. And of course it's not
appropriate for many identifiers, as they may issue IDs for things
that we don't intend to have in Wikidata, so we will never completely
cover them.

I should quickly plug the "expected completeness" property which is
really useful for identifiers - P2429 - as this can quickly show
whether something is a) completely on Wikidata; b) not complete yet
but eventually might be; or c) probably never will be. Not very widely
rolled out yet, though...

Andrew.


On 7 September 2017 at 19:51, Marco Fossati <fossati@spaziodati.eu> wrote:
> Hi everyone,
>
> As a data quality addict, I've been investigating the coverage of external
> identifiers linked to Wikidata items about people.
>
> Given the numbers on SQID [1] and some SPARQL queries [2, 3], it seems that
> even the second most used ID (VIAF) only covers *25%* of people items circa.
> Then, there is a long tail of IDs that are barely used at all.
>
> So here is my question:
> *which external identifiers deserve an effort to achieve exhaustive
> coverage?*
>
> Looking forward to your valuable feedback.
> Cheers,
>
> Marco
>
> [1] https://tools.wmflabs.org/sqid/#/browse?type=properties "Select
> datatype" set to "ExternalId", "Used for class" set to "human Q5"
> [2] total people: http://tinyurl.com/ybvcm5uw
> [3] people with a VIAF link: http://tinyurl.com/ya6dnpr7
>
> _______________________________________________
> Wikidata mailing list
> Wikidata@lists.wikimedia.org
> https://lists.wikimedia.org/mailman/listinfo/wikidata



--
- Andrew Gray
  andrew@generalist.org.uk

_______________________________________________
Wikidata mailing list
Wikidata@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata