W3C home > Mailing lists > Public > public-rdf-wg@w3.org > September 2012

Re: life without dataset semantics

From: Sandro Hawke <sandro@w3.org>
Date: Wed, 19 Sep 2012 09:36:03 -0400
Message-ID: <5059CA43.5000704@w3.org>
To: Steve Harris <steve.harris@garlik.com>
CC: "Peter F. Patel-Schneider" <pfpschneider@gmail.com>, David Wood <david@3roundstones.com>, "public-rdf-wg@w3.org Group WG" <public-rdf-wg@w3.org>
On 09/19/2012 05:37 AM, Steve Harris wrote:
> On 19 Sep 2012, at 00:22, Sandro Hawke wrote:
>
>> So, elsewhere you're proposing we not have dataset semantics.   I think I'm okay with that, if I can still do what I'm trying to do here.      What I'm not entirely clear on is how I can do this without any semantics....
>>
>> On 09/18/2012 02:10 PM, Peter F. Patel-Schneider wrote:
>>>> Here's a much better example, because it stays away from Web stuff:
>>>>
>>>>     <g1> eg:sendCorrectionsTo <mailto:sandro@w3.org>.
>>>>     <g1> { w3c:group35462 rdfs:label "SPARQL Working Group" }.
>>>>     <g2> eg:sendCorrectionsTo <mailto:ivan@w3.org>.
>>>>     <g2> { w3c:group44350 rdfs:label "RDFa Working Group" }.
>>>>
>>>>
>>>> There's an obvious meaning to the predicate eg:sendCorrectionsTo, but how do I express that meaning? Something like:
>>>>
>>>>     X eg:sendCorrectionsTo Y
>>>>
>>>>         Note: only meaningful inside a dataset which has a named graph with
>>>>         the name X.
>>>>
>>>>         Meaning: Y is a good email address for sending corrections to the
>>>>         information in the named graph X.
>>>>
>>>> Are you comfortable with that?
>>> I don't know if comfortable is the right word.  I don't have problems with anyone wanting to do that.
>> My question is really: do you think that definition/documentation means what I want it to and will work the way I want it to, if the RDF WG doesn't give Datasets any semantics?
>>
>> That is, if the WG doesn't say anything about graph-name URIs connecting to URIs as used in the default graph, can I just spell all that out (as above) in the documentation of my predicate?
> [slightly off-topic, sorry]
>
> I'm intrigued about this fixation with the default graph…

It comes from the (Web Architecture) idea that a data source needs to be 
usable given only its URL.  If you're given a thousand data sources, you 
just need to be given a thousand URLs.   This allows all sorts of nice 
scaling.    The alternative, where you're given a table of thousand 
records, each of which has various fields/properties, makes 
bootstrapping much harder.

Perhaps there's a way to get this property without giving the default 
graph special status, but I'm not sure what it might be.      I guess 
one could link to a named graph inside its dataset by using a fragment 
URL.    That might work....     (It would require particular wording in 
the trig media type registration to be fully spec compliant.)   It seems 
much more complicated.

> in practice we've not found it to be a useful place to store graph metadata, for a couple of reasons:

Well, I think you're not exchanging Datasets among diverse systems, so 
the above property doesn't matter much yet.
> 1) it assumes you'll only have one source of metadata - this is often not the case, we have metadata from the harvesting process, the IE process, and the currency checking process. They live in different graphs. If that isn't the assumption, then it assumes that you don't care where the metadata came from, which seems strange.

No, it doesn't assume that.   I'm thinking of the default graph as just 
the root of the metadata tree; given that, it's possible to link to the 
other metadata graphs, so folks can find them.   Without that root, how 
would another system know where to find those metadata graphs?

> 2) it more-or-less assumes you'll never update the graph (and hence metadata) otherwise it makes it very difficult to separate the metadata about one graph, from another. If your graph metadata is very, very regular,and you never change it's format over time then you could do it with a SPARQL Update, but that doesn't happen in real life very often.
>
(same response as above)
>> And if I can't, then what alternative do I have for sharing this kind of information structure?
>>
>>>   I can see that if this is the stance that someone wants to take with respect to named graphs, then one might want to have the relationship between IRIs and graphs work the way it works in the minimal semantics.
>>>
>>> However,  I don't think that everyone wants to have this connection.
>> What I'm asking for is an extremely weak connection; it's hard for me to see how it would do any harm, since it would only come into play when someone ask for it.
> But on the other hand, what good would it do?

It lets people exchange knowledge in datasets like they do in RDF, 
knowing only the URL, and using vocabulary terms they can look up if 
they don't already know them.   Datasets have the advantage over graphs 
in that they allow the triples to be separated, which is of course 
useful in managing data coming from multiple sources.

     -- Sandro

> - Steve
>
Received on Wednesday, 19 September 2012 13:36:13 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 16:25:51 GMT