Thursday, March 15, 2007

semantic web suffering

I'm getting a pretty good taste of why the semantic web hasn't exactly taken off today. I'm trying to use an OWL ontology produced by one of the world's leading universities in semantic web research to classify a couple RDF instances..

This would seem to be a straightforward task neh? Simply add the ontology to a reasoner, feed it the RDF containing the instances, and have a look at any inferred types it identifies.. Right?

Wrong!

A) The ontology utilizes qualified cardinality restrictions..
So, the current version of Jena (that I have) can't really handle it because this is OWL1.1
Pellet accepts these nasties (is mostly owl1.1. compliant I think) but ran for more than 20 minutes trying to classify it before I gave it up (on relatively new MacBook with 2gb ram)
Pellet has now been running for 20 minutes on our IBM P570..
and this is just an 80 class ontology!!!

B) To create the RDF instances that fit into the ontology framework, I had to either create them as RDF types of a qualified cardinality restriction (which Jena couldn't handle in OWL mode) or add multiple object property assertions that are apparently producing something invalid..

(from command line Pellet)
WARN [main] (KnowledgeBase.java:898) - Inconsistent ontology. Reason: Individual http://www.owl-ontologies.com/unnamed.owl#Afu3g10970 has more than 2 values for property http://www.owl-ontologies.com/unnamed.owl#containsDomain violating the cardinality restriction

I'm starting to feel like an idiot for a variety of reasons. I'm not used to having so much trouble making things like this work - its kind of my job.. and if its this difficult and time-consuming for some one who actually cares about it, how is the technology ever going to penetrate the great majority of the world that simply wants something to work? Perhaps I've been foolish in advocating the W3C's ideas for the semantic web.

Hopefully better news later this week

3 comments:

Richard said...

Most of the W3C's Semantic Web specs aren't really that complex—RDF and RDF Schema and SPARQL and GRDDL are all quite simple at their core. OWL is different in that you more or less need to be a logician to understand what's going on. But OWL 1.1 isn't even a W3C spec, it's a non-standardized work-in-progress by an outside group.

Arin said...

What you described looks like a problem in semiology: you may want to sort out the semiotics problem for your ontology

Benjamin Good said...

Hi Arin,
I did resolve my problems with some manipulations of the ontology and the representations used for the instances. I'm curious what exactly you mean by a semiotics problem.
Could you elaborate a little?