W3C home > Mailing lists > Public > public-media-annotation@w3.org > April 2011

RE: EBUCore and Eurovision NewsML-G2 in RDF

From: Evain, Jean-Pierre <evain@ebu.ch>
Date: Tue, 5 Apr 2011 09:56:26 +0200
To: 'Pierre-Antoine Champin' <pierre-antoine.champin@liris.cnrs.fr>
CC: "public-media-annotation@w3.org" <public-media-annotation@w3.org>, "ecm@list.ebu.ch" <ecm@list.ebu.ch>
Message-ID: <7D1656F54141C042A1B2556AE5237D60010ED4977A01@GVAMAIL.gva.ebu.ch>
Thanks pa, we are coming closer to the point ;-)  I really don't have much time left right now but I'll come back and challenge you with more questions soon.

Thanks again. Jean-Pierre

-----Original Message-----
From: Pierre-Antoine Champin [mailto:pierre-antoine.champin@liris.cnrs.fr] 
Sent: mardi, 5. avril 2011 09:47
To: Evain, Jean-Pierre
Cc: public-media-annotation@w3.org; ecm@list.ebu.ch
Subject: Re: EBUCore and Eurovision NewsML-G2 in RDF

comments inline

On 04/05/2011 09:24 AM, Evain, Jean-Pierre wrote:
> Hello Pierre-Antoine,
> 
> I followed the link...
> 
> There is a lot to read beyond the initial page but if I focus on this
> page only, and on your comment, I guess we can say it is about
> materialising inferences.  I agree that it is actually easier to get
> the inferences being published and in my scenario the user is
> actually more an aggregator whose role is to collect data and publish
> it with all the inferences allowing to do what MAWG was aiming at,
> i.e. cross domain queries.
> 
> If I take the example of ma-ont and EBUCore:
> 
> I have the object property 'features' in ma-ont equivalent to the
> object property 'features' in ebucore.
> 
> However, the range in m-aont is 'person' while the range in ebucore
> is 'contact' knowing that the class 'person' is declared equivalent
> to 'contact' too.
> 
> If the aggregator has collected datasets in the ma-ont and ebucore
> formats and reindexed the data with the mapping schema,


by "reindex", I assume you mean "assert all inferred"/"materialize the
inference"

> I am hoping
> that a SPARQL query in the form 'x features y' will e.g. for the same
> 'x' provide a list of 'y' made of persons and contacts (possibly
> overlapping, and / or complementary).

if the triples are there, any SPARQL engine should indeed retrieve them.

> Discussion: However, if an aggregator collects data including
> inferences and e.g. based on the same RDF schema, these data sets
> will be coherent but it may also be true that some properties have
> been instantiated in a set that would allow to develop richer
> inferences when combined with another data set.  This is maybe when
> option 2 of my scenario has some sense?

Ah, I see.

For me this is orthogonal to options 1 and 2 (as I understood them).
RDF/RDFS/OWL do not make a distinction between vocabularies and
instances... This is all a bunch of triples.

So if you import in your triple store the following RDF graphs

* a vocabulary V1
* a dataset D1 using V1
* a vocabulary V2
* a dataset D2 using V2
* a dataset D3 using V2
* a mapping M1 btw V1 and V2

you can either

* publish the whole set of triple and let users make inferences on them
(your original contribution being the mapping M1)

* run inferences on the triple store, and augment it with all the
materialized inferences so that users only have to query your triple store

Inferences can either be allowed by the mapping M1 (adding links between
terms from V1 and V2) or by the conjunction of the different datasets...
Again, from an RDF point of view, this is just inferences.

(of course, it depends on the inference engine you are using... but if
you stick e.g. to something like OWL-RL, you can run a rule engine until
saturation)

  pa


> 
> Anyway, I'll keep following the thread of pages that you pointed me
> to. That looks very informative. But after NAB ;-)
> 
> Regards,
> 
> Jean-Pierre
> 
> 
> -----Original Message----- From: Pierre-Antoine Champin
> [mailto:pierre-antoine.champin@liris.cnrs.fr] Sent: lundi, 4. avril
> 2011 16:05 To: Evain, Jean-Pierre Cc: public-media-annotation@w3.org;
> ecm@list.ebu.ch Subject: Re: EBUCore and Eurovision NewsML-G2 in RDF
> 
> Hi Jean-Pierre
> 
> from the two options you present, it seems to me that your question 
> boils down to
> 
> "should I materialize inferences or not?"
> 
> or in other terms
> 
> "should I compute all the inference once and for all and store
> them?"
> 
> where option 1 is "yes" and option 2 is "no".
> 
> Some elements of answers can be found here
> 
> http://patterns.dataincubator.org/book/ch04s07.html

> 
> -- and the short answer is "yes and no" :-)
> 
> pa
> 
> 
> 
> On 04/04/2011 09:55 AM, Evain, Jean-Pierre wrote:
>> Hello Pierre Antoine,
>> 
>> I'll try to go one step further in exploring the problem for which
>> I don't have a clear answer (if there is to be one ;-).
>> 
>> Maybe the easiest is to go through scenarios:
>> 
>> Scenario 1: A user develops a green-field application and decides
>> to use ma-ont. Descriptions / individuals are generated in this
>> format using the RDF provided in the ontology and stored in a
>> triple store (or else). The reasoned does the indexing following
>> the schema. Searches can be made using e.g. SPARQL.
>> 
>> The same user realises that there is a lot of data conforming to
>> one or more additional formats e.g EBUCore or NewsML-G2, both
>> expressed in RDF. He starts harvesting the triples available in
>> these formats and also wants to re-index this data from different
>> sources. The user therefore acquires the schema and uses its
>> reasoned to index the data.  This is primarily done on each set of
>> data e.g. EBUCore and NewsML-G2 RDF, all results being agin stored
>> e.g. in a triple store.
>> 
>> In order to combine this data and achieve richer query results.
>> The user either develop its own RDF mapping between the different 
>> ontologies or reuses an existing one (like one of the files I 
>> distributed)......
>> 
>> This is where is the dilemma:
>> 
>> Option 1: he reindexes all the data in its triple store
>> (containing ma-ont, EBUCore and NewsML-G2 triples) using only the
>> mapping RDF schema provided separately.
>> 
>> Option 2: he creates a super schema importing ma-ont, EBUCore and 
>> NewsML-G2 in RDF, plus the mapping RDF schema and uses its
>> reasoned to reindex all the data
>> 
>> Is option 1 better than or equivalent to option 2? Is there an 
>> another option? Etc.
>> 
>> Are there other scenarios that could allow addressing the problem 
>> more specifically ?
>> 
>> Best regards,
>> 
>> Jean-Pierre
>> 
>> 
>> 
>> 
>> -----Original Message----- From: Pierre-Antoine Champin 
>> [mailto:pierre-antoine.champin@liris.cnrs.fr] Sent: samedi, 2.
>> avril 2011 12:01 To: Evain, Jean-Pierre Cc:
>> public-media-annotation@w3.org; ecm@list.ebu.ch Subject: Re:
>> EBUCore and Eurovision NewsML-G2 in RDF
>> 
>> Hi Jean-Pierre,
>> 
>> On 03/30/2011 01:14 PM, Evain, Jean-Pierre wrote:
>>> (...) These two ontologies offer maximum compatibility with
>>> MA-ONT
>> 
>> very nice :)
>> 
>>> I have done the exercise to define the equivalence between the 
>>> classes and properties, which is really easy in the above 
>>> circumstances.  However, I am facing a dilemma. Should I declare 
>>> the equivalences in a separate 'mapping ontology' pointing to the
>>>  EBUCore, MA-ONT and NML-G2 ontologies as remote resources, or 
>>> 'import' them within their own namespace in one ontology 
>>> containing the equivalence declarations?
>> 
>> not sure what you mean by "pointing to X as a remote resource",
>> nor by "importint X in its own namespace"...
>> 
>> could you provide a short example of both approach?
>> 
>> pa
>> 
>> ----------------------------------------- 
>> ************************************************** This email and
>> any files transmitted with it are confidential and intended solely
>> for the use of the individual or entity to whom they are addressed.
>> If you have received this email in error, please notify the system 
>> manager. This footnote also confirms that this email message has
>> been swept by the mailgateway 
>> **************************************************
> 

Received on Tuesday, 5 April 2011 07:57:01 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:17:41 UTC