Nathan wrote:
Kingsley Idehen wrote:
Nathan wrote:
Hi All,
Apologies if this is the wrong place to ask questions about linked
data; however not sure where else to turn at the minute! and again
as it's quite a long list.
worth noting the following link for most of the following questions:
http://sameas.org/text?uri=http%3A%2F%2Fdbpedia.org%2Fresource%2FLondon
1] Let's say I'm writing an article about London, England; which one
of the many URI's do I reference that my data is "about"?
2] Would there be scope for a single globally unique identifier /
URI to represent "London, England"? one which rather than holding
information about London (like http://dbpedia.org/resource/London),
essentially held a set of sameas items which everyone could use when
publishing data "about" "London, England" (like the data at the
sameas.org link above).
3] If sameAs indicates that two URI references contain information
about the same thing; how do we assert that two URI's contain the
same information about the same thing (ie identical data)?
You don't want to assert that they have the same data. You are
asserting co-reference i.e. the URIs are about the same Entity. Thus,
you can then perform union style expansion from the co-reference URIs
to get a bigger picture of a given entity e.g., London, from a
variety of data sources.
Examples:
- About Me (compact) [1]
- About Me (expanded via explicit co-reference of the kind delivered
by owl:sameAs) [2]
- About Me (expanded via fuzzier co-reference via a rule that
asserts, in this context, that foaf:name is an inverse functional
property i.e., its values are in-direct identifiers) [3]
3a] as [3], mirrors are common on the net, us1.domain us2.domain
etc; each one containing the same information; as above how would
one indicate that the data is the same? considering that ...
- the data is identical, no way to inject in a "sameas" in to the rdf
RDFizer middleware can use custom (context specific) rules to make
their own assertions as part of the RDFization processing pipeline.
For instance, SPARQL is an effective rules language (head and body
just happen to be on a vertical as opposed to horizontal visual
plane), so it is possible for said engines to perform constrained
forward-chaining (with the generated triples written to a specific
graph that used in specific context).
- one 3rd party may reference the uri
http://us1.domain.com/something.rdf whilst another 3rd party
references http://us2.domain.com/something.rdf
both are the same data, but no correlation between the two exists
anywhere to say they are the same thing.
See comment above.
- it stands to reason that the ideal is a single endpoint and
mirrors behind the scenes without any http 30* redirects ever being
returned to the client, however this won't always be the case so
what syntax can we use in this scenario?
Single cannot exist.
Context is all that exists.
Within a given context (always inherently subjective) certain
assertions can be made about co-reference be it explicit (owl:sameAs)
or fuzzy (e.g. IFP based rules).
[4] Are there any conventions or guidelines for combining data and
resolving discrepancies? for instance to get all data about london
one would theoretically have to combine all the data from the uri's
referenced at (the sameas.org link aforementioned), but surely if
you combined all data together then you'd get both duplicates and
differences in the data.. which is fact etc.
[4a] Likewise with people - I have multiple social profiles all
about "me" but surely in the near future multiple URI's will each
represent #me; I think we can safely say that not all of these will
be linked with sameas, and further still which one should X person
use when referencing information about "me"?
[4b] Is there any method to mark which is the preferred source of
information (and verify it)? at the minute it seems like it would be
very simple to publish a vast amount of inaccurate data in triples
and it appears the current mentality would be to take it for granted
that the information IS fact.
DC vs ctag and FOAF
For RDFa we have ctag and maker; which to me seems very exact:
<span rel="ctag:means" resource="http://dbpedia.org/page/Washington"/>
<span rel="foaf:maker" resource="http://faviki.com/person/example#me"/>
but in dublin core we have the very loose
<span property="dc:subject">Washington</span>
<span property="dc:creator">Example</span>
I'm aware one can couple both dc and ctag/foaf in RDFa; but should
we be replacing dc values wherever possible with the more precise
ctag/foaf? (and indeed in our standard rdf data?)
A quick question about the usage of RDFa; previously I had always
envisioned RDFa documents to contain a lot of inline rdf markup; I'm
aware of the problems in picking up a term in the middle of a block
of text and wrapping it in the appropriate notation; however my
question is am I wrong in thinking this is the main use/advantage?
in most cases where I've sen XHTML+RDFa (like uriburner etc) it's
been more case of using RDFa to display human readable RDF; as
opposed to human targeted article with rfda embedded in-place /
in-line. Does anybody have any examples of a full RDFa demo site;
not just with the normal dc/foaf and tags but fully enriched with
detected semantic terms highlighted, linked and wrapped in rdfa,
inline..?
And finally any info on creating a set/document which comprises of
or includes / references items in other datasets? (I may really show
my newbie-ness here) - what I mean is say I'm making an RDFa page
about London, and in that I mention the population; I don't want to
have the population in document or in the rdf, I do however want to
link through to the triple which holds the population for london in
dbpedia or a geo set and have that in my rdfa. So where I could have:
(s–p–o)
london-population-7556900
I'd rather have:
london-population-{some link to dbpedia-owl:populationTotal value in
dbpedia's rdf for london)
Thus I'm saying that london's population is {found here} and it'd be
nice if it can also be pulled in and displayed through in an
XHTML+RDFa document by possibly
content="URI#dbpedia-owl:populationTotal" or suchlike.
Not sure if I explained that properly, perhaps just simply how do I
reference a single triple rather than a full rdf set; or am I way of
target?
Many Thanks in advance for any answers, comments etc & apologies
again if it's the wrong place to ask!
Nathan
I've used My person entity URI instead of "London", for maximum
effect i.e., lots of URIs associated with me etc..
Links:
1. http://tr.im/DoCA -- compact description (&sas=no implies no
"owl:sameAs" context rule)
2. http://tr.im/DoD4 - owl:sameAs expansion (&sas=yes implies
"owl:sameAs" smushing/meshing/expansion/explosion context on)
3. http://tr.im/DoIi -- show UI that provides holistic view of the
data space i.e., you can see via the indirect co-reference the effect
of an IFP rule re. foaf:name and foaf:mbox_sha1sum (note: there is a
bug I hit while writing this mail and you will most likely hit it if
you click on the IFP tab URIs)
4. http://tr.im/DoNv -- above using London from the larger data
corpus (8 Billion) at: http://lod.openlinksw.com (just visit the tabs
for the different co-reference URIs).
Thanks Kingsley!
something just clicked and half of my questions are now irrelevant; to
summarise my current understanding..
Let's say I'm doing the simplest report ever, where I want to display
the sentence "The population of X is Y" in XHTML+RDFa.
- where X is the current name of "London" and "Y" is the current
population, even if the name changes in 50 years to "new london" and
the population drops to 54
- and I want the data to be always up to date (or as up to date as
sources allow)
then all I need to do is:
1 - find a resource which holds rdf information about London
2 - SPARQL said resource to pick out only the name and population nodes
If using a SPARQL engine with inference rules capability plus the
ability to crawl within a SPARQL solution processing pipeline, you can
enable "owl:sameAs" inference context, and also ask the engine to follow
"owl:sameAs" links; thereby enabling you to get an expanded data set (on
the fly) to which the query pattern is applied, en route to final solution.
3 - (optional) assign a nice endpoint display the results of the query
as rdf
You don't display anything as "RDF" per se. you have data
representations based on the RDF model, and from these you can make
different presentations e.g. HTML or HTML+RDF (if you want to make your
presentation document a structured data data source based on the RDF model).
4 - XSLT transform the results in to an XHTML+RDFa document ["about"
URI for london], [by FOAF:Person/dc:creator me] where name and
population are injected in to X & Y respectively.
Sure, you can even use xslt in the SPARQL Protocol URL to achieve this goal.
that makes sense, and I'll assume the following:
to ensure info is always available I'd need to query (not sure what or
where here) to get all resources which describe city london, then use
all resources from above query as my source for steps 1-4 above; thus
should dbpedia die my report will still work!(?)
See comment above, re owl:sameAs and crawling withing SPARQL processing
pipeline. In addition to that you can cache you query results in you own
data space, and even apply cache invalidation schemes to that particular
space (remember HTTP gives you the mechanics for this gratis).
I guess that also advocates the decentralized nature of LOD and uses
of sameas, but on the other hand it suggests a need for a single point
of entry / initial search in to the "cloud"?
No single point of entry per se. I would think more in terms of
discovery and the degree of serendipity that the Web offers as its
structured linked data aspect gets denser.
and to ensure I've "got it"..
essentially I could write the following sentence in a document
"'Kingsley Idehen' wrote a post entitled 'name of post'"; where the
'name of post' is automatically injected at render time directly from
the rdf title of your post; so if you change the title, my sentence
remains accurate.
Yes :-)
All that's left is:
- my question regarding "combining data and resolving discrepancies"
(unless I find the answer upon closer analysis of the provided links);
Just look at the tabs in the links from my previous posts. The "indirect
co-reference" by nature of its essence, should reveal obvious errors,
since foaf:name alone cannot be a serious basis for co-reference (even
the fuzzier indirect variety).
- which are the preferred ontologies to use when trying to be very
specific about a subject (rather than dc.subject dc.creator etc which
are essentially free text based not URI identifier based)
Depends on what your describing. From my vantage point (or world view)
the foundation ontologies are :
1. FOAF
2. Dublin Core
3. SIOC
4. GoodRelations
5. Bibliographic Ontology
6. Music Ontology
7. SKOS
Regards & many thanks,
Nathan
--
Regards,
Kingsley Idehen Weblog: http://www.openlinksw.com/blog/~kidehen
President & CEO
OpenLink Software Web: http://www.openlinksw.com
