document.129

In the current literature of knowledge management and artificial intelligence, several
different approaches have been tried to the problem of developing domain
ontologies from scratch. All these approaches deal fundamentally with three
problems: (1) providing a collection of general terms describing classes and
relations to be employed in the description of the domain itself; (2) organizing the
terms into a taxonomy of the classes by the ISA relation; and (3) expressing in an
explicit way the constraints that make the ISA pairs meaningful. Though a number
of such approaches can be found, no systematic analysis of them exists which can
be used to understand the inspiring motivation, the applicability context, and the
structure of the approaches. In this chapter, we provide a framework for analyzing
the existing methodologies that compare them to a set of general criteria. In
particular, we obtain a classification based upon the direction of ontology
construction; bottom-up are those methodologies that start with some descriptions
of the domain and obtain a classification, while top-down ones start with an abstract
view of the domain itself, which is given 
a priori
. The resulting classification is
useful not only for theoretical purposes but also in the practice of deployment of
ontologies in Information Systems, since it provides a framework for choosing the
right methodology to be applied in the specific context, depending also on the
needs of the application itself.