<trackbot> Date: 19 January 2011

<cperey> Hi

-> http://www.w3.org/2010/POI/wiki/Drafts WG drafts

Discussion of id Primitive

-> http://www.w3.org/2010/POI/wiki/Core/Draft POI Core Draft

<cperey> I'll be signing off at half past

<scribe> scribe: Matt

matt: The ID we mentioned at the F2F only had to be unique to a particular system, not globally unique. I thought a URI would work here to get both, but there was a pushback.

alexh: I'd rather figure out what that means before I try to comment on it.
... I'm thinking like in XML where you can have an ID but it's not required.

<cperey> so the question is if an ID is REQUIRED or not?

<cperey> http://en.wikipedia.org/wiki/Uniform_Resource_Identifier

alexh: So, you aren't required to have an ID but it isn't required unless there is a reference to it.

matt: URIs give us a globally unique method of assigning IDs.

cperey: The wikipedia entry has a diagram that shows URIs above URLs and URNs.
... But we're not always talking about a Web environment.
... Is there still debate about whether an ID is required?
... Is it required or not?

alexh: I am interested in that too. I don't think that's an obvious need.
... And if we decide it is required, whether it's a URI or at a document level.
... I need to hear justification as to why each POI needs these things.

cperey: We have insufficient people to answer that question

<cperey> geopriv

cperey: If we had Henning here he would convince you

-> http://tools.ietf.org/wg/geopriv/ Geopriv docs

cperey: He would say all information needs to have an embedded classifier.
... Whether it's a POI or any other piece of information. It's one school of thought, but not the only one that advocates for these ids.
... And the Semantic Web is like that.

matt: Yep. And one thing to say is that the URI doesn't have to be http.

rsingh2: OGCs opinion would be that we need a unique ID too. I relate it to the IRS tax id.

cperey: Same philosophy at Geopriv.
... Have to be classified in some way.

alexh: One of the motivations for this is that some of these standards don't have IDs, and we consider that relieves a certain burden on generating the data.

cperey: If you do require it, then which system is it that we're all going to abide by.

rsingh2: Seems simple to me.

cperey: What classification system would you espouse?

rsingh2: At the most relaxed, just a string.
... Since it's generated by, or maybe the generators id plus a unique key for it.

cperey: That's about provenance, really.

rsingh2: No one wants to go to a central registry to get globally unique IDs. Combine own URI with some unique string.

matt: And that URI doesn't have to be dereferencable, just has to be unique.

cperey: Example: I've got a cup, I want to put it in a database and it contains coffee. When I generate that reference in the database something creates a unique identifier.
... It's stored in the database, and now the next time my sensor detects that object, then a pretty picture appears on the cup.
... Now, Alex's put the same cup in at the trade show. It gets a different unique identifier or different?

alexh: For me, in the AR context, if somebody describes that there is a cup out there, I expect to have some information on how to identify it uniquely. Perhaps a visual signature that the system may recognize.
... Once I find it, it's important for me to dereference it to get more information about it, but to find it, I don't need the unique ID.
... There was discussion about how to find these, the name, etc.
... So maybe information came along with it, and how to describe it, but it doesn't have an ID yet.

rsingh2: Any identifying information about this POI can change, except for the ID. A name, could have a spelling error you found it, but then someone wants to access that again and the name has changed.

alexh: What about this situation: two different databases are giving me a POI that everyone agrees is a POI, then I'll get two different IDs for the same POI.

rsingh2: I'd love to be stricter to say that we have a global id with a central registry.

alexh: So you're saying that the ID works within the domain of that ID system.

rsingh2: We're not saying globally unique yet, but just a unique ID in a system.
... It would be great to update in one place.

alexh: There's no unique marker on a coffee cup to distinguish it from yours.

cperey: Or we don't say anything about it, that the creator says information about it.

alexh: There may be two POIs, a picture of your dog and my dog that is on these coffee cups. Then it comes down to the descriptor of the object.

cperey: Seems to me that in a physical plane, the lat/lng/alt is the unique id for those that are fixed in space.

jacques_: No.

cperey: There may be other things at that lat/lng/alt?

rsingh2: The unique id is the thing you know will never change.

cperey: In a perfect fixed world, things that are fixed always have the same lat/lng/alt.
... So, the question then is why? How is this burden of adding the unique id and who do we get it from? Where does it come from? What is the benefit?

alexh: If the way you are identifying it shifts, then you need a way to track it, a unique ID to refer to it.
... I wanted to be sure this was the same POI, that was spelled differently, then I need a unique ID, until then, I don't.

cperey: I think we've got the benefit: disambiguation. But it's not necessary in all cases.

<Luca> Thanks Matt

rsingh2: If you want something, what's the burden on developers? Anyone serving out POIs has a database of them, and they have an internal way of tracking them. They'll be maintaining an ID, and it's not a big burden for them to include that information.
... It's hard to imagine it being a big burden for POI database people to do this.

alexh: So a POI gets created, and I introduce it, say it now exists. Now I have the burden of creating an ID.
... In some sense, I might use this datastructure to describe the POI, but I won't be providing an ID.

cperey: I've got to go, good discussion!

<cperey> and great scribbing!

alexh: It's clear to me the discussion, it's straight forward, we can use URIs, web address plus some sort of generated scheme. But I'm up in the air as to whether it's something to force.
... We need some guidance between the must and the can.
... I don't have an argument with requiring it, but I'm on the side of the fence that there is precedence for exchange systems that don't require it.

jacques_: If you use the ID of the guy who creates the POI, say a foaf ID, and you're using the dataformat with a local ID, then the guy who creates the POI and the POI id together, you have a likely global ID.
... We use some foaf elements, we identify who made the POI, and use the local ID for a unique POI ID.

alexh: That's perfectly reasonable to me, just whether it's required or not.
... Maybe we should table this.

rsingh2: Sounds like you're asking for a use case for a unique ID.

matt: action item?

alexh: Either Gary or Karl had to have this, so let's wait until we get them on the line and have another round on this.

matt: I'll email him privately about it.

Discussion of time Primitive

alexh: Looking at Google Earth, they timestamp things in KML. We had discussion that if there was no timestamp that it is permanent.

jacques_: Perhaps a bit more than a timestamp, but a start and end time?

alexh: If you know that information.
... A beginning and an end for historical things. For things that currently exist when it came into existence is useful.

matt: I think we could have more rich time stamps, that could apply to anything on the POI, not just it's existence.

alexh: To me, it's existence is what the timestamp is for. The rest is extensible data.

jacques_: I agree.

matt: But not built on the same primitive?

alexh: Looking at something like the Battle of 1826. You assign the start and end time to when that event happened.
... Nobody cares that it's put into the POI database on 7 Jan 2011.
... In that sense the timestamp is when it's valid.
... The timestamp could be 6-7pm every week.

matt: I guess I was getting at: should this be richer than just a begin/end rather than a simple timestamp.

rsingh2: Is this the same as the ID discussion?
... If you have a timestamp that isn't metadata, it becomes a way to identify the item. Combine the timestamp with the company ID you have the unique ID.

<jacques_> a timestamp for the browser to know if this POI has to be shown

alexh: What guarantee do you have that the timestamp will have granularity to guarantee it's unique.

rsingh2: Our specification would identify it.

alexh: I think that this is getting into an area where we as people trying to decide what a POI needs, can agree that it needs some rich timestamping, e.g. happy hours on Friday, but specifying that seems out of our purview. Somewhere out there there is a format that defines time.

<jacques_> a tmestamp for validity of the POI is different for timestamp related to the content of the pOI

rsingh2: ISO has a great definition of time, but I think we're hearing that we have great use cases, but none are mandatory.

alexh: Agree it wouldn't be mandatory. If it's completely metadata, then we have no standard to describe it.
... We probably do want to make some sort of element that may have different interpretations, but we at least have a way of describing time data.

<jacques_> no timestamp means the the browser will always show the POI

matt: I'm nervous to say we have different interpretations. I think we can agree we need to describe the basic building block though.

alexh: We have lots of things it could mean, creation, open hours, etc.

matt: I'm thinking the time building block could be applied to other primitives, e.g. the circus is here in July and here in January.

alexh: Good point, people are going to want to associate times with locations. At some point they have to decide whether it's two different POIs, or whether it's a change in database tracking.
... Looking at KML they have location with series of times. The other option is to make multiple POIs.
... Regardless, we do need that.

<jacques_> agree

alexh: For example if the POI is me, then someone might want to know how long Alex has been at his desk, when did he arrive there, etc.
... If you look on Foursquare I might be at a bar. If you look at the timestamp, you could say that's probably not where he is.
... Do we need two time stamps, or a time stamp associated with location.
... I would argue for it associated with location.

<jacques_> or a serie of time stamps?

alexh: But perhaps this gets to be a slippery slope: what if someone changes the name? Is there a timestamp for that? That's something I'm resistant to as well.

rsingh2: Me too.

matt: Why?

alexh: I expect the world is going to be full of this kind of data, that every nuance is going to be recorded and stored somewhere.
... By saying that I'm resistant to it, I'm not denying it, but I'm concerned if you start building it into a POI then it might become unwieldly.

rsingh2: It's a burden on developers without a clear advantage.

alexh: Something like tracks, it could be accumulated and make it available.

matt: How would you make it available?
... Separate POIs?

alexh: There is some precedence of people creating separate data elements that encapsulates the reference to the POI we're talking about and a time relation pairs. You can use that to create a path.
... It's data about the POI but not data within the POI.

matt: I guess I'm advocating for it to be available, but not required. I think the information is valuable to have access to in a standardized way, but doesn't have to be on every POI.

alexh: I can't argue against that.

F2F

matt: So far it looks like everyone is available on the 29-31 March dates. We're narrowing in on a location in Amsterdam.

This is scribe.perl Revision: 1.135  of Date: 2009/03/02 03:52:20  
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: RRSAgent_Text_Format (score 1.00)

Succeeded: s/unique POI/unique POI ID/
Found Scribe: Matt
Inferring ScribeNick: matt
Default Present: +3539149aaaa, Matt, +33.4.76.61.aabb, jacques, alexh, cperey, Raj, Luca
Present: cperey alexh jacques Luca
Regrets: Gary Karl Ronald Jens
Agenda: http://lists.w3.org/Archives/Member/member-poiwg/2011Jan/0013.html
Found Date: 19 Jan 2011
Guessing minutes URL: http://www.w3.org/2011/01/19-poiwg-minutes.html
People with action items: 

WARNING: Input appears to use implicit continuation lines.
You may need the "-implicitContinuations" option.