RE: Call for comments on your implementation experience. Bossam

Hi,

This is our experience of implementing Bossam rule engine and
applying it for OWL inference.

First, the overall story of implementation.
The inference engine we use, Bossam, is a production rule engine we
developed in this year. We tried to design Bossam so that it can be 
comfortably used for reasoning on the web. After finishing the
first phase of Bossam development, we started to write inference rules
for OWL. At first, we did not understand well the differences among
OWL species, but now we've just begun to understand. We're
reorganizing a rulebase for OWL. We're currently working on
OWL DL inference.

And here're some numbers:
- 1 month for understanding ontology and OWL *in general*. 
We had no prior knowledge on description logics, then. But we knew
about RDF, frames, FOL and AI in general. It was and still is difficult
for us, non-experts in logics and computational theory, to catch exactly
what the differences and meanings imposed on different OWL species
are. OWL AS&S is a notorious document among developers in here.
- 5 months for implementing the first (quasi-)stable version of Bossam
- 1 month for designing Buchingae, a rule language for Bossam, and
implementing a parser for it. We used ANTLR for writing syntax rules
and generating a parser.
- 1/4 month for implementing an OWL importing feature for Bossam.
We used Jena2 to produce RDF triples from OWL documents.
The triples are transformed into a series of facts, which are finally
loaded into Bossam's working memory. This was quite straight
and simple.
- 1/2 month for writing inference rules for OWL. We began from OWL
test cases one by one and wrote rules for each test case. We
soon realized that it was not a proper process of working out 
inference rules for OWL.
- 1/4 month for writing inference rules for OWL DL. Still far from
complete, and there're still many things to be learned. Currently,
Bossam has passed 50 tests out of 88 approved positive entailment
tests. There are 15 facts and 55 rules defined in the rulebase
for OWL DL inference. It took about 6 minutes to load W3C's
wine ontology and derive all the facts derivable by the rulebase.

Hope it helps.

Best,
Minsu

--------------------------------------------------
Minsu Jang
Senior Member of Engineering Staff
Business Knowledge Research Team
Electronics and Telecommunications Research Institute
Phone: +82-42-860-1250 Fax: +82-42-860-6790

Received on Wednesday, 19 November 2003 23:58:47 UTC