RE: Clarification required: BP6 "use HTTP URIs for spatial things"

And closing an important loop, _within_ the graph denoted
<https://g.co/kg/m/013qr8?_view=vertical-obstruction>

You will usually *not* find this URI used as the subject of an statements providing properties of a vertical obstruction (or navigation aid), but rather *will* find a bunch of statements with the subject

<https://g.co/kg/m/013qr8>

From: Cox, Simon (L&W, Clayton)
Sent: Friday, 2 September 2016 11:00 AM
To: 'Jeremy Tandy' <jeremy.tandy@gmail.com>; Rob Atkinson <rob@metalinkage.com.au>; janowicz@ucsb.edu; Joshua Lieberman <jlieberman@tumblingwalls.com>; Frans Knibbe <frans.knibbe@geodan.nl>
Cc: SDW WG Public List <public-sdw-wg@w3.org>
Subject: RE: Clarification required: BP6 "use HTTP URIs for spatial things"


Ø       <https://g.co/kg/m/013qr8> foaf:isPrimaryTopicOf <https://g.co/kg/m/013qr8?_view=vertical-obstruction>

With the corollary that
https://g.co/kg/m/013qr8?_view=vertical-obstruction

is the URI for a particular representation, possibly a graph.

From: Jeremy Tandy [mailto:jeremy.tandy@gmail.com]
Sent: Friday, 2 September 2016 1:45 AM
To: Rob Atkinson <rob@metalinkage.com.au<mailto:rob@metalinkage.com.au>>; janowicz@ucsb.edu<mailto:janowicz@ucsb.edu>; Joshua Lieberman <jlieberman@tumblingwalls.com<mailto:jlieberman@tumblingwalls.com>>; Frans Knibbe <frans.knibbe@geodan.nl<mailto:frans.knibbe@geodan.nl>>
Cc: SDW WG Public List <public-sdw-wg@w3.org<mailto:public-sdw-wg@w3.org>>
Subject: Re: Clarification required: BP6 "use HTTP URIs for spatial things"

Hi Rob - thanks for illustrating your concerns.

You've already voted +1 for using "indirect identifiers" for spatial things which helps me split out some of the issues you raise.

I think your concerns are about how URIs are actually resolved in practice. Clemens raised similar concerns in his earlier email [1].

> URI + some client-defined context  -> a resource that the server believes meets the context requirements

Yes. I'll admit to not following the 'reification' bit of your discussion completely, but this is how I see things working in practice. The user agent (client) identifies the Resource that they want (using the URI) and provides a bit of context for their preferences (commonly language and format). The server then determines what is the best information payload that it can provide in response ... and if the URI identified a physical entity such as Eddystone Lighthouse then obviously the HTTP response has to be an information resource describing Eddystone Lighthouse - we don't yet have the Star Trek teleporter technology to return concrete, steel and other physical atoms!

FWIW, the report from Geonovum's Geo4web-testbed Topic 3 provides some useful commentary on content negotiation strategies [2]

> So - if we have two FeatureTypes (Vert.Obstruction) and (Nav..) then we have two scenarios we need to cover:
>
> 1) U -> {R1, R2)
> 2) U1 -> R1, U2 -> R2, U1 = U2

In case #1 you raise valid concerns about the limitations of content negotiation. These concerns were also identified in Geo4web-testbed Topic 4: "Content negotiation based on media types insufficient" [3]. It says that content negotiation is "unable to support selection of different schemas (e.g. schema.org<http://schema.org> for the search engines vs DCAT for governmental data portals)".

In your stated example, how does one ask for a particular representation; R1 or R2?

BTW: I think that this is similar to the motivation that drives Erik Wilde's "profile" Link Relation Type proposal (RFC 6906) [4]

Imagine I'm using the Google Knowledge Graph ID for Eddystone Lighthouse <https://g.co/kg/m/013qr8> how could I say "please tell me about Eddystone Lighthouse seen as a vertical obstruction" (rather than as a maritime navigation aid) because that's what my application needs.

The Linked Data API (LDA) provides a solution in the form of "views" [5]. One can configure an LDA server to expose named views; these are accessed using a URL query parameter (e.g. "?_view="myNamedView"). The LDA server uses a SPARQL Construct statement to select only those properties that are relevant to the named view when creating the HTTP response.

What I don't see in the LDA is any canonical mechanism to advertise or describe the set of named views available from a particular server so that a client application knows what they can request - but it's possible I missed this. I think that you described these as "reified bindings of U to R1...n".

> In summary - as an implementer who now understands the probIem I have a small number of questions:
>
> 1) in the latter case is BP to declare U1 owl:sameAs U2?  If not - what is the predicate ?

I think <owl:sameAs> is correct.

> 2) what is the appropriate predicate for the  U predicate R1 ?

Taking the practice from LDA, they (or at least Epimorpics' "ELDA" implementation [6]) use <foaf:isPrimaryTopicOf>, so for my lighthouse example, I might state:

     <https://g.co/kg/m/013qr8> foaf:isPrimaryTopicOf <https://g.co/kg/m/013qr8?_view=vertical-obstruction>

And there's the inverse <foaf:primaryTopic> if you want to assert in the other direction.

(I'm a bit rusty on the Linked Data API - I hope I got that right!)

> 3) How do I share the context (i.e. those reified bindings of U to R1...n) so a client can understand them

I think that this is an area where best practice doesn't exist - and accordingly we should flag this as a potential interoperability bear-trap.

Jeremy

PS: I think that this discussion is highly relevant to the BP doc, but suspect that it better fits in §10.6 Spatial Data Access [7] rather than within the BP about identifiers.

[1]: https://lists.w3.org/Archives/Public/public-sdw-wg/2016Aug/0152.html

[2]: https://github.com/geo4web-testbed/topic3/wiki/Content-negotiation

[3]: http://geo4web-testbed.github.io/topic4/#h.n0gkernttzw0

[4]: https://tools.ietf.org/html/rfc6906

[5]: https://github.com/UKGovLD/linked-data-api/blob/wiki/API_Viewing_Resources.md#specialised-viewers

[6]: http://www.epimorphics.com/web/tools/elda.html

[7]:  http://w3c.github.io/sdw/bp/#bp-exposing-via-api


On Wed, 31 Aug 2016 at 23:24 Rob Atkinson <rob@metalinkage.com.au<mailto:rob@metalinkage.com.au>> wrote:

Important we have this in depth discussion and reach consensus - because its impossible to expect a wider audience to arrive at a usefully common view without guidance.

This stuff is out there, so its not a theoretical discussion. but some of the practices conflate things in a way that makes data hard to integrate.

There are a couple of practical implications I havent seen fully addressed, yet would need to be highlighted IMHO:

 One mechanism in widespread use is that the URI for the 'thing'  redirects to a URL for a resource about the thing.  Lets call
 these U & R.   What is important to recognise here is that the relationship between U and R is mediated by conneg... In the case of the UK Linked Data example, it may be further mediated by additonal parameters within the Web architecture.

So the information content is always URI + some client-defined context  -> a resource that the server believes meets the context requirements.

At the very least U -> { R1, R2,R3...}

-> can be seen as some type of property relationship - so this behaviour can be automated or also stated in a useful way

U hasRepresentation R1

or perhaps reified

{ subject U, object R1, predicate hasRepresentation, type aFeatureType, label "X', dct:hasFormat "image/svg" }

The reified case makes it obvious that the "client defined context" can be matched to the properties of the relationship - however this logic is just as likely to be implemented in the conneg or API layer, thats a contract between client and server which is part of the broader Web architecture - for now let us focus on whether the functionality required can be met by the available information...

now if  U1 -> R and U2 -> R then both U1 and U2 may have the same set of properties in this case

X owl:SameAs Y  states that the properties of X apply to Y and vice versa. There are other implications here as well


he value constraint owl:hasValue<http://www.w3.org/TR/2004/REC-owl-semantics-20040210/#owl_hasValue> is a built-in OWL property that links a restriction class to a value V, which can be either an individual<https://www.w3.org/TR/owl-ref/#Individual> or a data value<https://www.w3.org/TR/owl-ref/#Datatype>. A restriction containing a owl:hasValue constraint describes a class of all individuals for which the property concerned has at least one value semantically equal to V (it may have other values as well).

NOTE: for datatypes "semantically equal" means that the lexical representation of the literals maps to the same value. For individuals it means that they either have the same URI reference or are defined as being the same individual (see owl:sameAs<https://www.w3.org/TR/owl-ref/#sameAs-def>).
So I beleive they key thing here is to provide guidance to those unfamiliar with the semantics underpinnings to not break things and given them a simple way forward.

So - if we have two FeatureTypes (Vert.Obstruction) and (Nav..) then we have two scenarios we need to cover:

1) U -> {R1, R2)
2) U1 -> R1, U2 -> R2, U1 = U2

In summary - as an implementer who now understands the probIem I have a small number of questions:

1) in the latter case is BP to declare U1 owl:sameAs U2?  If not - what is the predicate ?

2) what is the appropriate predicate for the  U predicate R1 ?

3) How do I share the context (i.e. those reified bindings of U to R1...n) so a client can understand them

If we cannot nail down the terms to use now, then I believe we need to state in the BP that a mechanism needs to be chosen to address these specific concerns according to the principles of re-use - but I also suggest we flag this as a critical "must solve" issue for the communities we would suggest pointing users to.

Rob


On Thu, 1 Sep 2016 at 03:13 Krzysztof Janowicz <janowicz@ucsb.edu<mailto:janowicz@ucsb.edu>> wrote:
Thanks for the clarification.


On 08/31/2016 09:22 AM, Jeremy Tandy wrote:
@josh

> Is the triangle spatial data or a graphic with drawing instructions that assumes a certain technology? If 100 people print out the SVG, there is really nothing to indicate that the underlying entity is the same on each piece of paper, just that the same instructions were used, unless we want to get into trademark issues.

This seems to be getting away from the main topic. Unless you object, can I pull us back?

@Krzysztof

Apologies if my terminology is confusing.

I was trying to say that <owl:sameAs> indicates that two identifiers (URIs, in this case <http://example.com/sar/features/vo/EDY> and <http://example.org/maritime/navaid/2650253>) refer to the same entity (Eddystone Lighthouse). You said it much better than me.

The term "representation" was drawn from @josh's email text; in which he meant "Eddystone Lighthouse seen as a vertical obstruction" and "Eddystone Lighthouse seen as a maritime navigation aid".

> there is no need for such a class [whose members are both vertical obstructions and maritime navigation aids] (which you can define if you really want to, but it could lead to a combinatorial explosion)

I agree. This is what I've seen with Linked Data implementations - which means that "sameRealWorldEntityAs" is not required.

Hmmm. I hope I'm not confusing myself and everyone else.

Jeremy

On Wed, 31 Aug 2016 at 17:02 Joshua Lieberman <jlieberman@tumblingwalls.com<mailto:jlieberman@tumblingwalls.com>> wrote:
On Aug 31, 2016, at 10:20 AM, Frans Knibbe <frans.knibbe@geodan.nl<mailto:frans.knibbe@geodan.nl>> wrote:



On 31 August 2016 at 13:42, Joshua Lieberman <jlieberman@tumblingwalls.com<mailto:jlieberman@tumblingwalls.com>> wrote:
If we are asserting that spatial data on the Web is "always" feature data that represents a real world entity, then yes, we don't have the general Web "is it or isn't it physical" ambiguity and can assume that a feature data identifier also and indirectly identifies the feature.

I hope we can broaden that assumption, that the assertion still holds even if we are not talking about feature data representing real world entities.

Let's look at a border case: I am drawing a triangle in Inkscape and I save it as a *.svg file. I publish the file on the web, so it has a URI. Now I would say the triangle is a spatial thing (not sure if it counts as a real world entity, but I hope we can leave the idea of 'real world' out of definitions anyway). The SVG object in the file is the geometry describing the spatial thing. I think that only if we understand the SVG file to be the spatial thing we get into trouble. I might want to state that the file has a certain size and that the triangle has a certain area. It would be funny if I used the same URI for both statements. So I would need to have a different URI for my triangle. Could that be all?

Is the triangle spatial data or a graphic with drawing instructions that assumes a certain technology? If 100 people print out the SVG, there is really nothing to indicate that the underlying entity is the same on each piece of paper, just that the same instructions were used, unless we want to get into trademark issues.


That still leaves a gap in expressing whether two feature data entities represent the same real world entity. Perhaps we need a "sameFeatureAs" predicate to address this.

Yes, that is what the Subject equality<http://w3c.github.io/sdw/UseCases/SDWUseCasesAndRequirements.html#SubjectEquality> requirement is about. So the BP document is expected to say something about that.

Regards,
Frans


Josh
Joshua Lieberman, Ph.D.
Principal, Tumbling Walls Consultancy
Tel/Direct: +1 617-431-6431<tel:%2B1%20617-431-6431>
jlieberman@tumblingwalls.com<mailto:jlieberman@tumblingwalls.com>

On Aug 31, 2016, at 07:29, Frans Knibbe <frans.knibbe@geodan.nl<mailto:frans.knibbe@geodan.nl>> wrote:
Hello,

As stated before, I don't think the httpRange-14 problem exists in our domain of discourse. I think (and hope) that confusion can only occur when the things that are described are digital things, or things that can be transmitted over a computer network, like web pages or mail boxes. It seems to me that spatial things are never that type of thing. Therefore there is no reason to take precautions against possible confusion.

That probably means +1.

Greetings,
Frans



On 31 August 2016 at 09:50, Jeremy Tandy <jeremy.tandy@gmail.com<mailto:jeremy.tandy@gmail.com>> wrote:
Thanks Rob & Clemens ...

On Wed, 31 Aug 2016 at 08:30, Clemens Portele <portele@interactive-instruments.de<mailto:portele@interactive-instruments.de>> wrote:
+1


On 30 August 2016 at 10:10:26, Jeremy Tandy (jeremy.tandy@gmail.com<mailto:jeremy.tandy@gmail.com>) wrote:
Hi. It would be good to close this issue out & include our collective recommendation in the BP doc working draft.

PROPOSAL: SDW working group recommends use of "indirect identifiers" for spatial things

... I'll start the voting.

+1

Jeremy

(BTW, to make sense of the PROPOSAL you'll need to read the email thread)

On Fri, 26 Aug 2016 at 10:12 Linda van den Brink <l.vandenbrink@geonovum.nl<mailto:l.vandenbrink@geonovum.nl>> wrote:
So… do we agree we can recommend indirect identifiers, or do we try to fix the issue with getting the correct identifier as Rob describes?

While waiting for this I’ve updated the issue and the text referring to the issue in BP6.

Van: Rob Atkinson [mailto:rob@metalinkage.com.au<mailto:rob@metalinkage.com.au>]
Verzonden: woensdag 24 augustus 2016 13:56
Aan: Jeremy Tandy; Phil Archer; Linda van den Brink; Bill Roberts

CC: SDW WG Public List
Onderwerp: Re: Clarification required: BP6 "use HTTP URIs for spatial things"

Hi

Agree this is a real concern - people cant be blamed for doing the obvious, if dumb, thing..

I think we should take note of best practice in the HTML world - which is often to include a citable link to a resource in the rendered view.  Or a "share" or something similar. We can also put fairly explicit annotation in machine-readable code - stating that the resource is about the URI - and even notes saying when citing this resource use the URI....

I'd also like to see browsers evolve to offer you the original link or the redirected when cutting and pasting - how hard can it be!

Maybe we can get Ed to ask around Google Chrome team for suggestions on how best to handle this :-)

Rob



On Wed, 24 Aug 2016 at 18:27 Jeremy Tandy <jeremy.tandy@gmail.com<mailto:jeremy.tandy@gmail.com>> wrote:
Yes, I think so ... And we should do so if we are recommending "indirect identification".

Jeremy
On Wed, 24 Aug 2016 at 09:24, Phil Archer <phila@w3.org<mailto:phila@w3.org>> wrote:
Bill's comments also made me think about some of the classic arguments,
such as that a lake doesn't have a last updated date and isn't 435KB
big. Which are true, however, that kind of metadata generally comes from
the server, i.e. the HTTP layer. That's an over simplification but the
point is that it is relatively easy to avoid deliberately creating
misleading metadata - metadata about the doc rather than the thing it
describes - and it's also generally easy to avoid looking for that metadata.

Is there scope for some BP advice there?

Phil.

On 24/08/2016 08:25, Jeremy Tandy wrote:
> Thanks Linda. More clear examples where being "correct" (in terms of
> avoiding uri collisions by using two distinct uris) is making things worse
> because users take the wrong one!
>
> So, as a WG, are we content to recommend this "indirect identification"
> pattern where thing & info resource identifiers are conflated?
>
> Bill has added some good points about how to avoid impacts of uri
> collision- by using the (dataset) metadata to talk about licenses and
> creators for the information ...
> On Wed, 24 Aug 2016 at 07:52, Linda van den Brink <l.vandenbrink@geonovum.nl<mailto:l.vandenbrink@geonovum.nl>>
> wrote:
>
>> Experience from the Netherlands: we have the id/doc pattern in our URI
>> strategy, based on the Cool URIs note [8] and the ISA study on persistent
>> identifiers [9].
>>
>>
>>
>> That being said, same as Bill I also notice data users getting confused
>> and generally using the /doc/  URI as that is the one they can copy from
>> their browser address bar. This is not only casual confusion but also ends
>> up in published information resources.
>>
>>
>>
>> You see this, for example, all over the CB-NL which is a vocabulary for
>> the building sector and contains links to other Dutch standards such as
>> IMGeo, an information model and vocabulary for large scale topography. E.g.
>> the CB-NL concept of ‘Gebouw’ (Building) [10]  links to two IMGeo concepts
>> ‘Pand’ (building part) and ‘Overig Bouwwerk’ (other construction) using
>> their /doc/ URIs. If you click on Pand (which doesn’t have its own landing
>> page in CB-NL so I can’t include the link) you will see it includes the
>> /doc/  URI as the identifier of Pand.
>>
>>
>>
>> This is an example where it occurs in vocabularies, but I also see it
>> happen with identifiers for data instances.
>>
>>
>>
>> [8]: https://www.w3.org/TR/cooluris/

>>
>> [9]:
>> https://joinup.ec.europa.eu/sites/default/files/D7.1.3%20-%20Study%20on%20persistent%20URIs_0.pdf

>> 10: http://ont.cbnl.org/cb/def/Gebouw

>>
>>
>>
>> Linda
>>
>>
>>
>> *Van:* Jeremy Tandy [mailto:jeremy.tandy@gmail.com<mailto:jeremy.tandy@gmail.com>]
>> *Verzonden:* dinsdag 23 augustus 2016 20:57
>> *Aan:* Bill Roberts
>> *CC:* SDW WG Public List
>> *Onderwerp:* Re: Clarification required: BP6 "use HTTP URIs for spatial
>> things"
>>
>>
>>
>> Thanks Bill. Sounds very coherent ... I hoped for some responses such as
>> this based on practical experience. Jeremy
>>
>> On Tue, 23 Aug 2016 at 19:41, Bill Roberts <bill@swirrl.com<mailto:bill@swirrl.com>> wrote:
>>
>> ah Jeremy, you are a brave man to poke the sleeping beast of httpRange-14.
>>
>>
>>
>> But I'll get my thoughts in early, then I can tune out of the ensuing mail
>> avalanche :-)
>>
>>
>>
>> When publishing Linked Data about places we (at Swirrl) generally do the
>> id/doc fandango, but to be honest I think data users either don't notice,
>> or they get confused by it.  In the applications we are working with (and I
>> acknowledge that others may have different applications and different
>> experiences), it wouldn't cause any problems to have a single URI, the 'id'
>> URI if you like.  We just don't find a need to say anything about the /doc/
>> URI.  If we were starting again, I'd probably ditch the /doc/ and the 303
>> and rely on context and a little bit of documentation to make it clear what
>> we mean.
>>
>>
>>
>> The place where we find a need to talk about creators and licences and
>> modified dates is in metadata about datasets where a dataset might be a
>> collection of information about a bunch of places - and we treat datasets
>> as an 'information resource'. If someone requests a dataset URI we return a
>> status code of 200 and the dataset metadata as the response.  That metadata
>> includes info on where to get all the contents of the dataset if you want
>> that.
>>
>>
>>
>> By the way, though it's sensible and consistent, I find that the implied
>> and parallel property stuff makes it more rather than less complicated.
>>
>>
>>
>> Bill
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> On 23 August 2016 at 17:37, Jeremy Tandy <jeremy.tandy@gmail.com<mailto:jeremy.tandy@gmail.com>> wrote:
>>
>> All-
>>
>>
>>
>> Linda has done a great job of consolidating the best practices are use of
>> identifiers. We have just one [1] now.
>>
>>
>>
>> Reading though just now, it occurred to me that there's still an open
>> issue about identifier assignment ...
>>
>>
>>
>> W3C's Architecture of the World Wide Web constraint "URIs identify a
>> single resource" [2] asserts "Assign distinct URIs to distinct resources"
>> in order to avoid URI collisions [2a] which "often imposes a cost in
>> communication due to the effort required to resolve ambiguities".
>> Discussions from earlier years in UK Gov Linked Data working group (and
>> elsewhere) concluded that the "real world thing" and "information resource
>> that describes the real world thing" are separate resources. I think this
>> is based on a (purist?) view when working with RDF of needing to be totally
>> clear on "what's the subject" of each triple ... the thing or the document.
>> This manifests as URIs with `id` or `doc` included somewhere to distinguish
>> between the resources and some RDF triples to clarify that the doc resource
>> is talking about the thing resource etc..
>>
>>
>>
>> (dangerously close to "httpRange-14" [3] here ... let's avoid that bear
>> trap)
>>
>>
>>
>> Jeni Tennison's "URLs in Data Primer" draft TAG note captures this
>> practice in §5.3 "Publishing data" [4]:
>>
>>
>>
>> ```
>>
>> Publishers can help enable more accurate merging of data from different
>> sites if they support URLs for each entity
>> <https://www.w3.org/TR/urls-in-data/#dfn-entity> they or other sites may
>> wish to describe, separate from the landing pages
>> <https://www.w3.org/TR/urls-in-data/#dfn-landing-page> or records
>> <https://www.w3.org/TR/urls-in-data/#dfn-record> that they publish.
>>
>> ```
>>
>>
>>
>> Yet Architecture of the World Wide Web §2.2.3 "Indirect identification"
>> [5] notes that:
>>
>>
>>
>> ```
>>
>> To say that the URI "mailto:nadia@example.com<mailto:nadia@example.com>" identifies both an
>> Internet mailbox and Nadia, the person, introduces a URI collision.
>> However, we can use the URI to indirectly identify Nadia. Identifiers are
>> commonly used in this way.
>>
>> ```
>>
>>
>>
>> This is consistent with what I recall TimBL saying at TPAC-2015 in regards
>> to Vcard; come the finish, no one really cares to distinguish between the
>> thing and its associated information resource.
>>
>>
>>
>> ... And in most cases, one can use context to determine whether a
>> statement concerns the thing or the information resource. In those cases
>> where you can't, "URLs in Data Primer" suggests some mechanisms to mitigate
>> such confusion [6][7].
>>
>>
>>
>> I think that in our SDW WG discussion we have concluded that we _are_
>> content to use "indirect identification" - e.g. that we use URIs that
>> conflate the thing and document resource.
>>
>>
>>
>> Please can we confirm this? Assuming that indirect identification is
>> "approved" as best practice, then it seems prudent to add a note to the BP
>> document saying "don't worry about distinguishing between thing and
>> resource; indirect identification is fine" (etc.)
>>
>>
>>
>> Thanks, Jeremy
>>
>>
>>
>> [1]: http://w3c.github.io/sdw/bp/#globally-unique-ids

>>
>> [2]: https://www.w3.org/TR/webarch/#pr-uri-collision

>>
>> [2a]: https://www.w3.org/TR/webarch/#URI-collision

>>
>> [3]: https://www.w3.org/2001/tag/group/track/issues/14

>>
>> [4]: https://www.w3.org/TR/urls-in-data/#publishing-data

>>
>> [5]: https://www.w3.org/TR/webarch/#indirect-identification

>>
>