Re: Using RDB for a large OWL-DL database.

Bijan Parsia wrote:
> Just out of curiosity, what fragment of OWL does your translation 
> handle? I believe LUMB turns out to be in hornSHIQ:

Good question. I'm not able to say that in two words.
The algorithm first makes some subsumption inferences (currently on the 
basis of explicit subClassOf, oneOf, intersectionOf and unionOf 
axioms/descriptions).
Classification is done with regard to almost all OWL axioms and 
descriptions, by generating a DB view for each description/axiom on a 
triple table and combining these views as needed. In case of mutual 
recursion, there is currently some approximation used for computation of 
an initial class extend, which is refined by a second run. We didn't use 
some fix point algorithm due to performance reasons and for many 
practical scenarios this doesn't seem to be needed.
Non-unique names assumption turns out to be problematic, OWLDB does not 
check if certain individuals are the same. But I guess that won't be 
very problematic, when an individual is listed twice (for 
classes/properties of course it's a bigger deal).
I think it's a bit hard to compare the OWLDB approach to classic DL 
reasoners, since only subsumption and classification are supported, both 
with a different set of (in some cases only partially) supported OWL 
descriptions/axioms.
Let me know if you want to have a glance at the paper I'm currently 
writing once I have a first draft (might take 2-3 more weeks).

Cheers,

Sören