Re: agenda: 25 March RDF Data Access WG teleconference

Forwarded message 1

• From: Eric Prud'hommeaux <eric@w3.org>
• Date: Thu, 18 Mar 2004 08:21:09 -0500
• Subject: EP-1 - EP-4 including promised general metadata query
• To: public-rdf-dawg@w3.org
• Message-ID: <20040318132109.GB20106@w3.org>
• X-Archived-At: http://www.w3.org/mid/20040318132109.GB20106@w3.org

Here are some use cases from Annotea. I've labeled them with
discriptive names, as well as a short moniker. At the end I've loosely
tied this back to the general metadata requests we discussed in the
teleconf 14 hours ago. I'll post some more about tracking
attributions/provenance later.

ns rdf=<http://www.w3.org/1999/02/22-rdf-syntax-ns#>
ns a=<http://www.w3.org/2000/10/annotation-ns#>
ns t=<http://www.w3.org/2001/03/thread#>
ns http=<http://www.w3.org/1999/xx/http#>
ns dc0=<http://purl.org/dc/elements/1.0/>
ns dc1=<http://purl.org/dc/elements/1.1/>

EP-1
InputAnnotationQuery: Client1 posts a document to a server that
accepts (and agrees to host), amongst other things, a:Annotations. The
document contains an a:Annotation which a:annotates
http://example.com/doc1.html. The server distinguishes graphs that it
is willing to host by looking for a minimal subset of assertions in
the posted document:

ask (
 ?annotation rdf:type    a:Annotation.
 ?annotation a:annotates ?annotates.
 ?annotation a:context   ?context.
 ?annotation a:body      ?body)
collect (?annotation ?body)
----------------------------------------------------------------------

EP-2
AnnotatesQuery: Client2 asks about a page (for instance, the one that
Client1 has annotated). The server gives back all the properties and
values for any property coming from a node which also annotates doc1.

ask
(?annotation rdf:type    a:Annotation.
 ?annotation a:annotates <http://example.com/doc1.html>.
 ?annotation ?p          ?o)
collect (?annotation ?p ?o)
----------------------------------------------------------------------

EP-3
SubjectQuery: Client asks for a hosted object directly by name.

ask
(<http://example.com/annot1> ?p ?o)
collect (?p ?o)
----------------------------------------------------------------------

EP-4
FatAnnotationQuery: Client asks for a certains set of properties about
a hosted object. The ( tripleA || tripleB ) expressivity is used to
make efficient queries of graphs that mix the Dublin Core 1.0 and 1.1
namespaces:

ask
(<http://example.com/annot1>  rdf:type    a:Annotation.
 <http://example.com/annot1>  a:annotates ?annotates.
 <http://example.com/annot1>  a:context   ?context.
 ( <http://example.com/annot1>  dc0:creator ?creator ||
   <http://example.com/annot1>  dc1:creator ?creator )
 ?creator                     a:E-mail    ?email.
 ?creator                     a:name      ?name.
 <http://example.com/annot1>  a:ceated    ?created.
 ( <http://example.com/annot1>  dc0:date    ?date || 
   <http://example.com/annot1>  dc1:date    ?date )
 <http://example.com/annot1>  a:body           ?body.
 ?body                        http:Body        ?bodyData.
 ?body                        http:ContentType ?contentType)
collect (?annotation ?body)
----------------------------------------------------------------------

EP-2 (AnnotatesQuery) demonstrates one style of getting general
metadata about an object. The query is a simple (albeit with variable
predicats) graph query -- it does not assume any omniscience on the
part of the server. A form closer to today's discussion could be:

ask
(<http://example.com/foo> ?p ?o)
collect (?annotation ?p ?o)

This assumes that all the properties needed to determine what would be
"interesting" to know about foo are one directly attached. For
instance, it would be bad practice have a bookmark attached by an
rdfs:seeAlso. Using a b:bookmarkedBy property would give the queryer a
chance to know immediately if it's the sort of data that it's
interested in persuing.

For efficiencies sake, people may start to flatten their data
structures so that they will all be loaded when the client asks for
one level of properties away from foo. This would be unfortunate as
we'd see useful data structures like

<foo> :address <a1>.
<a1>  :street "Elm".
<a1>  :number 123.

decomposed into

<foo>  :address_street "Elm".
<foo>  :address_number 123.

eliminating some of the intuitive structure and data hiding utility of
having an address record.

The variable predicate approach does not address everything that a
well-educated server might provide when asked for "stuff about foo",
but I think that the remaining bits could be left to higher layers
like popular query profiles:

ask
(<http://example.com/foo> :browserProfile ?o)
collect (?annotation ?p ?o)

If we wanted an approach like this we'd have to make sure that the
server was not limited by the query lanuage specification to only
returning properties with foo in the subject and :browserProfile in
the predicate -- that it was free to return some annotations and some
ISBN info and whether it goes well with a cabernet...
-- 
-eric

office: +81.466.49.1170 W3C, Keio Research Institute at SFC,
                        Shonan Fujisawa Campus, Keio University,
                        5322 Endo, Fujisawa, Kanagawa 252-8520
                        JAPAN
        +1.617.258.5741 NE43-344, MIT, Cambridge, MA 02144 USA
cell:   +1.857.222.5741 (does not work in Asia)

(eric@w3.org)
Feel free to forward this message to any list for any purpose other than
email address distribution.