Information Technology Reference
In-Depth Information
13.2
Modeling of Medical Concepts
In this section, we discuss the notion of a natural concept and the creation of compu-
tational models for natural concepts. First, we describe natural concepts and place
our description in the context of Dr. Sadegh-Zadeh's categorization of “classical
concepts.” Second, we discuss concept modeling in the context of knowledge rep-
resentation. Next, we focus on medical concepts and their characteristics such as
context-dependency, changeability and imprecision.
Since our ultimate goal is to create computational data models for data mining,
we approach the “concept” definition from a representational perspective. Accord-
ingly, we view “concept” as a principle of classification. Furthermore, we approach
the concept definition from a cognitive perspective. First, we ask two fundamental
questions: “How do people classify objects into categories?” and “How do people
mentally represent categories?” The answers to these questions are fundamental
for the creation of computational models. The findings from cognitive psychology
about human categorization have demonstrated that category learning and classifi-
cation in the real world are different from the creation and classification of artifi-
cial categories (e.g., mathematical categories) [1, 14, 15]. Furthermore, cognitive
psychology describes three approaches to the mental representation of natural cat-
egories:
classical
,
prototype
,and
exemplar
[12, 14]. These three categories are
described below since the distinction between them is critical for the creation of
appropriate computational models in data mining.
•
Classical Approach.
In the classical approach, objects are grouped based on
their properties. Objects are either a member of a category or not, and all ob-
jects have equal membership in a category. A category can be represented by a
set of rules, which can be evaluated as true (object belongs to the category) or
false (object does not belong to the category).
•
Prototype Approach.
In the prototype approach, the members are more or less
typical of the category; in other words, they belong to the category to a certain
degree. The prototype of the category usually represents the central tendency
of the category and may be defined as the “average” of all the members of the
category. Thus, in the prototype approach, a category is based on a “prototype,”
which exists as an ideal member of a category, and the other members of the
same category may share some of the features with the ideal member [1].
•
Exemplar Approach.
In the exemplar approach, all exemplars of a category are
stored in memory, and a new instance is classified based on its similarity to all
prior exemplars [10]. This representation requires specification of a similarity
measure, as well as storage and retrieval of multiple exemplars. In the exemplar-
based representation, the category is defined by all exemplars belonging to a
given category.
