On 1/3/09 01:30, Giovanni Tummarello wrote:
congrats and kudos to all those who've made this happen. I think the
cloud diagrams are proving a very compelling visual for people who
don't care about nerdy detail but understand the idea of interlinked
datasets.
Yes they're great for handwaving if the audience has never seen it,
otherwise its likely counterproductive
The problem is that LOD has been stuck here 2 years really now, not a
single advance not a single application (of the LOD model, not of the
data, the data is obviously useful and expressing in RDF is also
starting to be seen as useful) .
Well, it *is* all about the data. Don't forget that! If the cloud
diagrams only serve to remind people that many many datasets overlap in
scope, and can be aggregated into larger units wherever they mention
common objects and use common vocabulary, all is well. We don't need
intelligent mobile agents for this to pay off. Just nice big databases
and good old fashioned code.
LOD is an elaboration and improvement on the original linking model we
had in FOAF (back before RDFCore when the RDF spec was vague on some key
points, like how many URIs a thing could have, how to model
same-thing-ness, ...). The main reason for rdfweb (as I called it
originally) was discovery. In 2000, there was basically no RDF in the
public Web, apart from some half-hearted bits of Dublin Core. No search
engines did anything with it then (vs today, with Yahoo, Google, Yandex,
Nutch, Sindice, SWSE et al.). So having an information linking model for
RDF was important: it meant we could pretty much find all the RDF in the
public Web by starting at one FOAF file and crawling. I think LOD has
similar value today, but the pressure to have a hypertext-based
discovery model is somewhat reduced. Partly because dataset-level
information is available (eg. URL templates for LiveJournal, or VOID for
LOD sites), and these tell coders how they can get their hands on huge
chunks of data. But also because there are more aggregators and lookup
tools.
Even if most apps work only with a single dataset, linking is
worthwhile. It reduces the degree of coupling between app and dataset,
by increasing the commonalities between datasets. And it's a nice hook
for crawlers, who can then expose different aggregate views back as more
bubbles.
That the bubbles continue to grown is however a sociological interesting
phenomen :-)
Nothing wrong with sociology! :)
I think as SKOS gets rolled out more seriously, linkage by topic (eg.
LCSH, Dbpedia, ...) will become worth its own custom visualisation...
On the positive side, i recently reviewed some work by someone who has
a very interesting way to create a diagram which actually helps by
showing which queries can be asked. Too bad you wont see it in action
at ESWC because the demo paper was "not up to the springer standards for
legibility", according to some other reviewer.
The problem here imho is that too many people have forgotten that it is
the "Semantic Web project", and instead treat "Semantic Web" as the name
for a research field, or for a hypothetical future version of the Web
that may never exist.
cheers,
Dan