Re: Time for quintuples?

Hi everybody,

Am 24.09.2010 14:11, schrieb Nathan:
> Toby Inkster wrote:
>> On Fri, 24 Sep 2010 01:12:45 +0200
>> Bob Ferris <zazi@elbklang.net> wrote:
>>
>>> However, if I have for example three "shortcut relations" (e.g. a
>>> skill, an expertise and and interest) in a graph, which have all the
>>> same topic, I can't match them to the related cognitive
>>> characteristic statement (as you can also see in the original
>>> example[1]).
>>
>> What's wrong with this?
>>
>> ex:APerson
>> cco:skill <http://dbpedia.org/resource/Football_(soccer)> .
>>
>> ex:CC1
>> a cco:CognitiveCharacteristic ;
>> cco:agent ex:APerson ; #added this
>> cco:characteristic cco:skill ; #added this
>> cco:topic <http://dbpedia.org/resource/Football_(soccer)>;
>> wo:weight [
>> a wo:Weight ;
>> wo:weight_value 6.0 ;
>> wo:scale ex:AScale
>> ];
>> cco:activity
>> <http://sw.opencyc.org/concept/Mx4rwJRiEpwpEbGdrcN5Y29ycA> .
>>
>> The following N3 rule would allow you to infer the shortcut property
>> from the longer form:
>>
>> {
>> ?cc
>> cco:agent ?s ;
>> cco:characteristic ?p ;
>> cco:topic ?o .
>> }
>> => { ?s ?p ?o . } .
>
> nice, +1 for Toby's approach from me, and likewise using n3 for all
> you're named graph needs.

Yes, I proposed this idea before also on the FOAF dev mailing list (see 
[1]):

"- introduce a further property with a domain of
cco:CognitiveCharacteristic, which directs to the applied cognitive
characteristic, e.g. ex:CC1 cco:applied_cognitive_characteristic cco:skill
	-> then we might somehow fall back to the "standard" 		
	reification approach (subject relation -> cco:agent/cco:habit; 	
	predicate relation -> cco:applied_cognitive_characteristic;
	object relation -> cco:topic)"

However, it might have the mentioned drawbacks ;)

With the quad based approach, one would "bind" the reification statement 
directly to its "shortcut relation".
Furthermore, what would happen, if you like to describe a cognitive 
pattern of the same category, e.g. a skill, and with the same topic, 
e.g. soccer, which should be related with two activities, e.g. goal 
scoring and goal keeping* (which should have different weightings etc.). 
Then you will need two cognitive characteristics descriptions, which are 
"bound" to their related "shortcut relations", or?

I'm getting more and more the opinion that we need a separation between:
- 'internal context', which should represent "semantically related 
information" to the triple (here identified as reification statement)
- 'external context', which represent "semantically unrelated 
information" to the triple (here identified by the Named Graph 
entailment and the description of the Named Graph)
as it was a result of Tolle's research (see "Understanding Data by their 
Context Using RDF"[2] for example, or [3], because the design of 
Triplesets was also inspired by this "separation").

Furthermore, the same idea was probably similar applied by the approach 
of the metaK framework (see "Querying for Provenance, Trust, Uncertainty 
and other Meta Knowledge in RDF"[4]), where they called the 'external 
context' 'meta knowledge' and introduce a separation between knowledge 
and meta knowledge (with separate interpretation of meta knowledge 
properties etc.). I really like the proposal, which is described in [4], 
because they are delivering a good explanation of their intended 
semantics and how they would integrated their approach in the existing 
Semantic Web framework.
Although, this proposal might have some drawbacks in scaling to a Linked 
Data scenario, because one have to define the intended meta knowledge 
properties (that mark them somehow) and their corresponding 
interpretation functions for query processing (this is currently done 
via their Java framework). Maybe a Semantic Graph based description 
might be good, so that other information consumer can also use this 
knowledge for query processing.
However, what would happen, if one have to mix different meta knowledge 
property definitions via a semantic federation framework, e.g. in one 
source a property is assigned as meta knowledge property and in another 
one not, or the interpretation functions of a meta knowledge property 
are different.
Furthermore, can't there also be a fall-back into a ranked result 
recommendation, if the filtering and/or ranking conditions of a query 
wouldn't be fully fulfilled and the exact search result would hence be 
empty? For instance, they drew an example that a more recent source 
would outperform an older, but what will happen if the newer one has a 
lower certainty?

Cheers,


Bob


*it might not be the best example, but hopefully good enough for 
explanation ;)


[1] 
http://lists.foaf-project.org/pipermail/foaf-dev/2010-September/010422.html
[2] 
http://www.dbis.informatik.uni-frankfurt.de/~tolle/Publications/2004/AISTA04.pdf
[3] http://www.ontotext.com/ordi/ORDI_SG/ORDI_SG_Specification.pdf
[4] 
http://www.uni-koblenz.de/~staab/Research/Publications/2009/jws-meta-knowledge.pdf

Received on Friday, 24 September 2010 13:35:09 UTC