[Pellet-users] Ontology stalls Pellet Redux

[Taowei David Wang]

So there are a couple of things.  There was a mistake in my previous
analysis of the lkif ontologies.  To make things easier, I had merged the
14 ontologies into 1, but in doing so, I accidentally clubbered (merged) 5
classes and a property (because they shared the same 'name' aside from the URI).
This faulty ontology can be accessed in [1].  Since then, I have made
another version that's correct (as far as I can tell) in [2].

There are still interesting things (from performance profiling
perspective) about [1]:
1) From Protege 4.0 Alpha, Fact++ took 104 minutes to classify.
2) From Protege 4.0 Alpha, Pellet took around 2 minutes to classify.

In [2] neither reasoner was able to perform classification from Protege
(seems to 'stall'), though CPU certainly kept working hard after many
hours.

There are many more classes in [2] Pellet could not perform satisfiability
check within 5 seconds (151 of them).  And there are only 17 known
unsatisfiability classes:

Valid_Rule Transaction Termination Argument2 Public_Body
Continuation Rule Code Trade Legislative_Body Decision
Mandate Initiation Assignment Delegation Act_of_Law
Regulation

There are only a small number of classes that can be profiled (37) in 5
seconds.  So the clubbered classes were making the ontology that much
harder. They are:

Argument, Assumption, Atom, and Exception (2).

The clubbered property is

overlap

Though this type of analyses is suggestive and useful, it's not
conclusive.  We still don't know exactly what axioms are causing the
bottleneck, or how to determine the dependencies of 'hardness' (if any),
or what choices were made in this particular profile run, or if we can
give an estimate of performance prior to running the tableau.

These are research questions I am currently facing.  I am glad that there
are interests from the community, and I hope to bring the tools and
services to a reasonable state before I release it.  But hopefully soon :)

Cheers,

David
-----

[1] http://www.mindswap.org/~tw7/work/profiling/others/lkif-all.owl
[2] http://www.mindswap.org/~tw7/work/profiling/others/lkif-all-correct.owl