Search Mailing List Archives
[protege-discussion] Ontology for enterprise systems
perisbrodsky2011 at u.northwestern.edu
Wed Mar 24 10:00:35 PDT 2010
This is mostly a generic ontology question, though Protégé may be part of
the answer. Please refer me to a more appropriate forum, if you know of
I am new to ontology for computing, as of several weeks ago. My perspective
is as a designer and builder of enterprise systems for organizations in the
health care sector, using a model-driven approach.
In my designs, I favor heeding the model-view-controller pattern, versus a
more "pure" object-oriented approach that allows any object class to have
behaviors. What this means is that in the model there is a clear separation
between objects containing persistent business state, but very little
behavior (entity objects) and transient object containing purely business
logic (command objects). One of the motivations for this separation is to
allow a distributed deployment of business objects, but be able to control
and allocate processor resources to the compute-intensive commands. Another
is to make commands the clear focus of automated testing.
The result of this is that, to me, an enterprise persistence model looks a
lot like an ontology. Much of the business data exist solely for the
purpose of driving the behavior of the business logic, whose purpose is to
solve problems in the enterprise domain.
The attraction of ontology to me, at this point, is that it purports to
separate knowledge from "operational" data. This would be a big benefit for
real-world deployment, because it would pave the way to sharing the KB
between the production instance of the system, and other instances used for
validation, training, etc.
The problem is, how do we decide what data are operational? The ontologies
I've seen include classes that seem operational. For instance, the
newspaper ontology, a Janitor class is explicitly excluded (per the Ontology
101 tutorial), because it's not needed to answer competency questions. Does
that make it make it operational? Is that the test? On the other hand,
Reporter is an ontology class, but it seems that instances of Reporter
should be operational. But instances are knowledge if they form part of a
competency question, or its answers. So where do we draw the line?
For a clinical decision support system, most of the decisions it makes
involve a patient and all that is clinically known about the patient. So a
clinical ontology must include class Patient, and many other associated
ones, and so its KB must include all Patient instances, right? So what's
the difference between and EMR and a KRS, if anything? There seem to be no
enterprise data that are not part of domain knowledge, in that they may all
potentially be involved in solving problems.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the protege-discussion