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

On Sep 27, 2006, at 7:40 AM, Jinsung Seol wrote:

> Hello, All!
>
> When I am thinking OWL, I cannot help worrying that inferencing a  
> new concept or fact from complex

Let's grant "complex", i.e., that you are talking about fairly  
complex axioms (perhaps involving disjunction and transitivity).

> and verbose

Actually, OWL (and I'll speak only of the DL part) is pretty terse.  
In SHOIN you can easily encode relational structures that are  
exponential in the size of the encoding. That's what makes reasoning  
with OWL DL hard.

> OWL informal

Uh...you meant "formal", right?

> descriptions is more difficult than finding relationship from  
> Relational Database,

I don't know what this means. Do you mean evaluating queries?

> because of the concept of open world.

Eh..I wouldn't think that's the main point. That you latch onto it in  
this manner indicates you aren't getting what's really going on.

> I already know OWL is based on RDF and XML,

No. OWL DL is based on SHOIN(D). It  can be serialized in RDF which  
can then be serialized in RDF/XML. Big difference. I recommend the  
description logic handbook as a starting place for understandign  
what's going on.

> but for a large amount of database it could be efficient to use  
> Relational Database as DL-based database,

There is work on combining DL reasoning with database techniques.  
Some keywords to help your googling: InstanceStore, KAON2, DLP,  
hornSHIQ, Abox Reduction (see this years DL workshop), DL Lite.

> if the transformation between OWL-DL and RDB is feasible and the  
> inference in the RDB is enable.

Google the keywords above.

> It is because nowadays RDB can be compatible with Web access.

This has nothing to do with the above problem. At all.

> Now, I have tested some OWL validator, also RDF validator which all  
> are treating the description as low level material such as XML, RDF.

I don't know what you mean here.

> Therefore, no semantic problem were not returned.

Or here.

> If the transition is feasible and efficient, inference from RDB can  
> be consistent and more, SQL can leverage to enable semantic  
> validation by using a nested SQL query, number calculation, text  
> search function, and the more.

Of course if you can reduce a particular OWL KB to a relational db,  
you can use the relational db to do OWL reasoning (if the reduction  
is sound). However, that alone might not give you any gain in  
performance and might actually be much worse (if the reduction  
involves a blow up, for example).

The KAON2 papers might help your understanding quite a bit. Check out  
Boris' thesis.

> How about this idea?

Various refined aspects of it have been tried and are still be  
researched.

> Is it valuable and valid in point of efficiency and integrity with  
> Description Logic?

It's debatable. There is certainly no silver bullet in this  
area...different techniques have different tradeoffs. The main focus  
has been on dealing with large ABoxes (which makes sense given the  
data orientation of databases). Complex or just large TBox reasoning  
has not fared well at all. (Though there's been no serious  
effort....however, I don't really see how it would go; Large ABox  
store might be the closest).

To sum up, it's definitely not nuts to try to use a database in some  
way, but I would strongly recommend you examine the existing rich  
literature to get a good sense of exactly what the problems/ 
challenges and promises are. At the moment, your grasp seems a little  
shaky (though some of this could be language issues).

Cheers,
Bijan.

Received on Wednesday, 27 September 2006 09:05:59 UTC