Re: Performance issues with OWL Reasoners

>>>>> "KV" == Kashyap, Vipul <VKASHYAP1@PARTNERS.ORG> writes:

  KV> OWL reasoners support two types of reasoning:

  KV> 1. ABox reasoning (reasoning about instance data). Scalability
  KV> here is being achieved here by leveraging relational database
  KV> technology (which is acknowledged to be scalable) and mapping
  KV> OWL instance reasoning operations to appropriate SQL queries on
  KV> the underlying data store.

I may be wrong here, but as far as I know the expressivity of OWL-DL,
for example, is too different from that of RDBMS for this to work
completely. I am not enough of an expert to know if this sort of
mapping is possible at all or whether it just cannot be done
efficiently. 

Having said that there is a similar approach, which uses RDBMS. For
example, the instance store (http://instancestore.man.ac.uk) which I
was briefly involved with (before the backend got to hard for my poor
brain!), uses a metaschema backend. Queries are not made by mapping to
SQL, but using SQL and reasoner queries together. 


  KV> 2. TBox reasoning scalability is a challenge, especially at the
  KV>    scale of 100s of
  KV> thousands of classes found in medical ontologies. Would love to
  KV> hear >From DL experts on this issue.

Again, as far as I understand, the complexity of T-Box and A-Box
reasoning for logics such as that underlying OWL-DL are not that
different (i.e. they are both terrible!), so the issues are much the
same. 

There is not general answer to the size of a T-box you can reason
over. If the T-Box is a simple asserted hierarchy, you can build a
pretty large ontology (certainly 10's of thousands) and reason over
it -- the reasoning in this case being simple. If you start using lots
of more complex expressions then you can limit yourself a lot more. 

This paper for example, managed to get the Gene Ontology and, I think,
all of GOA into a DL form and reason over it in a, er, reasonable
amount of time. So scalability to 10's of thousands of T-box and 100's
of thousands of A-Box's is possible. 

http://www.cs.man.ac.uk/~dturi/papers/instancestore2.pdf

The DL reasoners are much better than they used to be -- in the good
old days, when the world was young, you could get most DL reasoners
to eat your CPU on a 10 term ontology. Nowadays, it's fairly hard to
do this. 

Phil




-- 
Phillip Lord,                           Phone: +44 (0) 191 222 7827
Lecturer in Bioinformatics,             Email: phillip.lord@newcastle.ac.uk
School of Computing Science,            http://homepages.cs.ncl.ac.uk/phillip.lord
Newcastle University,                   Claremont Tower, Room 909
NE1 7RU

Received on Thursday, 14 September 2006 17:26:29 UTC