On 06.04.2015 22:02, Markus Krötzsch wrote:
Dear Sebastian,
Using OWL is surely a nice idea when the semantics is appropriate (i.e., where you want Open-World entailment, not constraints) and here the
Possibly misleading typo: I meant "where", not "here" ;-) -- Markus
expressiveness is enough. This is much more difficult, however, than one might at first think it is. For a simple example, the common Wikidata constraint that a property is /symmetric/ can not be expressed in OWL. The reason is that, in order to represent statements with references and qualifiers in OWL (i.e., in RDF), one needs to introduce auxiliary individuals for statements. I discussed some of these limitations of OWL in my keynote at the "OWL: Experiences and Directions" workshop 2012, but it seems the slides are not on the web site. I will try if I can track them down and publish them somewhere.
Maybe you already have observed these limitations yourself? I was not sure from your email (and the linked documents) how exactly you envision the use of OWL. One thing that is clear is that OWL does cannot be used on Wikidata directly, but only on an RDF version of it. For this reason, you should also have a look at the (many) ongoing discussions about the final details of this RDF model. You can find related issue reports on Phabricator. I think it is also fairly safe to base your work on the published RDF export (see our paper at ISWC 2014): there will be changes, but the basic structural aspects that matter for creating OWL statements will most likely be the same. The paper also contains some discussion of how current Wikidata constraints can be mapped to OWL (which of course is not the semantics that constraints have).
Maybe I should explain this again in detail, since some of these issues do not seem to be completely clear to the Wikidata community right now. For example, you can see things like Wikidata's "instance of" (P31) being declared to be "equivalent" (P1628) to rdf:type. Of course, this "equivalence" is only an informal notion that refers to the common ideas of classification that are embodied in both "properties" (note that they are both called "properties" but that there are fundamental differences between RDF properties and Wikidata properties -- again, they are closely related in spirit but not in a precise formal way). In particular, there is no semantic framework where P31 and rdf:type coexist, so it does not make sense to declare them "equivalent" in any stronger way. The best we can do is to translate Wikidata data into RDF, but after this translation, there is no "P31" any more: instead, there are several RDF properties that are used together to encode P31 statements, and none of these RDF properties is "equivalent" to rdf:type.
OWL is part of the RDF world, and it only has meaning in this context -- you cannot apply OWL to Wikidata contents directly. You can certainly apply OWL to RDF exported from Wikidata. However, it you want the resulting conclusions to be "first class" statements in the Wikidata world (as you seem to suggest), then you need to use an RDF encoding that faithfully captures all data in Wikidata. This is the reason why OWL can express the symmetry of RDF properties, but not the symmetry of Wikidata properties.
Best regards,
Markus
On 03.04.2015 11:16, Sebastian Burgstaller wrote:
Hello all,
Wikidata consists of millions of single data items, which is great. In order to facilitate modeling the interactions between the single items, we hereby suggest using OWL based ontologies (http://en.wikipedia.org/wiki/Web_Ontology_Language).
We think that using ontologies brings several advantages: -Looking at an ontology (could collaboratively be generated e.g. on webprotege.stanford.edu http://webprotege.stanford.edu) gives a very clear overview of how data is interconnected. This would allow for modeling of even very large and/or complex interactions. -Layouting a data integration project in an ontology first, before really integrating data into WD facilitates property proposal, as a ontology with its properties could first be designed and then the ontology with all its properties and classes could be generated as a whole. -Data could be queried/exported from WD based on an ontology by simply selecting the whole or parts of an ontology.
This approach has been suggested and discussed by Benjamin Good, Elvira Mitraka, Andra Wagmeester, Andrew Su and me. As an example, we put together draft properties for gene disease interactions, which allows for WD community discussion of this apporach. A preliminary version can be found here: https://www.wikidata.org/wiki/User:ProteinBoxBot/GeneDiseaseIteraction_Discu...
Best regards,
Sebastian
Wikidata-l mailing list Wikidata-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata-l