Completeness of classification results -- has anyone else noticed something similar to this and has an explanation ?

[songweihong]

I met the following problems, and hope to get some valuable advice !

I generated the Galen-Heart-n2 and Galen-EL-n1 ontologies by
updating some concepts beginning with lower case to nominal related
concepts on the original Galen-Heart and Galen-EL ontologies. And then
I classified them with Hermit and CB. However, Hermit got less
subsumption pairs than CB(direct subsumption) in both ontologies. For
Galen-Heart-n2, Hermit missed 6 pairs of direct subsumption(running
less than 3 minutes); and for Galen-EL-n1 Hermit missed 113 pairs
(running 45 hours).

However, I did satisfiability test with some of the missing pairs,
Hermit could get correct results.

I tested them in two ways:
1. Using Hermit source code, i.e., CommandLine.java in Hermit
project (the newest version of hermit1.3.6, 2012-02-22 version)
configuration was as follows:
config.blockingStrategyType =
Configuration.BlockingStrategyType.SIMPLE_CORE;
config.directBlockingType =
Configuration.DirectBlockingType.PAIR_WISE;
config.existentialStrategyType =
Configuration.ExistentialStrategyType.INDIVIDUAL_REUSE;
Reasoner hermit=new Reasoner(config,ontology);

-c -v3 -o<result path> <source path>

2. Using Hermit.jar in command line:
java -jar HermiT.jar -c -v3 --block-strategy=core --block-
match=pairwise --expansion-strategy=reuse <source path> --
output=<result path>

In the above configuration,, may I know whether Hermit should get
complete classification results ? If it should, why didn't Hermit get
complete results comparing with CB in these two ontologies ? Is it
possible that my extracted ontologies had problems ? Or my process
using Hermit had problems ?

I put these two ontologies, their classification results as well as
their difference on Hermit and CB in the bug report attachment(buglist
issue11). Could somebody please kindly have a look at them and give me
some feedback ? Thanks in advance !