[OSM-talk] Potential bot tasks relating to Wikidata errors

[Mateusz Konieczny]

13 lip 2022, 18:26 od andy at pigsonthewing.org.uk:

> On Wed, 13 Jul 2022 at 13:01, Mateusz Konieczny via talk
> <talk at openstreetmap.org> wrote:
>
>> 10 lip 2022, 16:14 od andy at pigsonthewing.org.uk:
>>
>>> On Sun, 10 Jul 2022 at 13:57, Mateusz Konieczny via talk
>>> <talk at openstreetmap.org> wrote:
>>>
>>>> Based on my own similar efforts: Wikidata has large scale issues
>>>> with it classificaton and I would recommend manual tool-assisted changes.
>>>>
>>
>>
>>> Relevant example?
>>>
>>
>> small sample:
>>
> I explicitly asked for a /relevant/ example. None of those you have
> given are relevant to the kinds of issues highlighted by the tool
> under discussion, or the proposed fixes to them; nor do they preclude
> fixes by bot.
>
In general you cannot assume that Wikidata classification is accurate,
so anything making automatic edits based on it is not acceptable in OSM.

Things like
- USS Niagara museum ship is classified as "group of humans"- all objects marked as canals are classified as "non-physical entity"- University of San Francisco is classified as an action
are enough to demonstrate that any bot relying on such source will cause problems.

Very low classification accuracy is relevant.

>> Overall, Wikidata classification system is not allowing to
>> reliably answer questions such as "is this an event" or "is it a physical object"
>> or "is it ship or group of humans" or "is it physical or non-physical entity".
>>
>
> Yes, it is.
>
No it is not (at least I failed to find one after sinking way too much time).

How one may check whether given wikidata item describes event?
(in way that is not listing https://www.wikidata.org/wiki/Q2859225 
and https://www.wikidata.org/wiki/Q49493599 as events
and lists Q134301 Q3182723 Q663435 as events?)

Maybe I am using wrong method, but Los Angeles Aqueduct and
community garden in New York City kept getting classified as events 

How one may check whether given wikidata item non-physical entity?
(in way that is not listing https://www.wikidata.org/wiki/ <https://www.wikidata.org/wiki/Q450876>Q57838673
https://www.wikidata.org/wiki/Q589884 https://www.wikidata.org/wiki/Q672804
https://www.wikidata.org/wiki/Q5734420 https://www.wikidata.org/wiki/Q30593659
https://www.wikidata.org/wiki/Q75320653  as non-physical entity)

Right now  all items listed here (cemetery, castle, artwork, garden)
are indirectly classified as objects that exists outside physical reality.

Anything fully automatic relying on source of that accuracy is not acceptable
for deployment in OpenStreetMap.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk/attachments/20220713/aa497ba6/attachment-0001.htm>