Re: ISSUE-1: Inferencing and current databases

On Mon, Apr 13, 2015 at 10:00 AM, Holger Knublauch <holger@topquadrant.com>
wrote:

> In order to make some progress on the question "what inferencing can or
> must be used", I started a wiki page with links to the reasoning support of
> various popular RDF/SPARQL databases
>
>     https://www.w3.org/2014/data-shapes/wiki/Entailment_Support
>
> Maybe some more knowledgeable people can provide additional details where
> needed, and add databases that I forgot.
>
> While many (but not all) database seem to support RDFS, I tried to
> understand the set up procedures for graph-level inferencing, which are
> sometimes quite involving (e.g. how to tell the system which class
> definitions to use for an instances graph). A problem that I see is that
> once inferencing is activated per graph, it seems to be impossible to
> distinguish asserted from inferred triples. This means that in a typical
> scenario for constraint checking, if someone edits an instance on a form,
> the form may also display inferred triples that in fact cannot be edited.
>
> So I have no idea how we could realistically rely on database-level
> inferencing at this stage.
>
> I would also like to hear from others, how many people have actually
> activated any of those inferencing modes in their deployed systems, and
> what the performance impact is. For example, Dimitris, is the dbpedia set
> up using inferencing?
>

I delayed too much the reply on this but it is related to the discussion we
had yesterday and the action on Arthur & Holger.
In DBpedia we perform type(-only) inferencing during extraction time.
A simplified workflow is: using the infobox-to-ontology mappings we assign
a type to a resource but when we write the type statements we transitively
get all super types defined in the DBpedia ontology and write the
additional type statements.
This makes querying or statistics about e.g. dbo:Person(s) or dbo:Place(s)
much easier and closer to the user expectations.
Although the DBpedia ontology has rdfs:subPropertyOf relations we do not
perform similar inferencing on the properties.

Best,
Dimitris


>
> Thanks,
> Holger
>
>
>


-- 
Dimitris Kontokostas
Department of Computer Science, University of Leipzig & DBpedia Association
Projects: http://dbpedia.org, http://http://aligned-project.eu
Homepage:http://aksw.org/DimitrisKontokostas
Research Group: http://aksw.org