Information Technology Reference
In-Depth Information
V
−
V
i
j
d
=
or
(5.1)
ij
max{
V V
,
}
i
j
V
V
V
where
,
are two possible values for feature
.
j
(2) The similarity among categorical features.
The similar degree for categorical features is generally identified into two
types: the simple method is to compare the corresponding values of the
features. If the two are identical, the similar degree between the features is 0;
otherwise, is considered to be 1. Other methods may incorporate some
differential ranking into the definitions of similarity according to the detailed
relations among the feature values, rather than the simple 0-or-1 division. The
former is actually similarity in quality, i.e. the division between to be or not to
be, while the latter is similarity in quantity, the further discrimination among
the features. Generally speaking, the former is common and suitable for all
kinds of situations; the latter is needed to be predefined and related to domain
knowledge, thus serves special purpose. The above two kinds of similarity
assessment methods have different application domains of their own.
(3) The similarity among ordered features.
Ordered features are the features lying between the numeric and the
categorical ones, thus the similarity degrees for them lie between the quality
and quantity assessments too. When the values for a feature are in order, we
can assign different similarity degrees to different segments of the whole line
of possible values. Compared with categorical features, the ordered features
have more regular similarity assessment rules. Suppose the whole line of
values can be divided into n segments, then the similarity between feature
values in segments i and j can be defined as
1
.
| i
−
j |
−
n
Features of different types can be transformed into each other. A feature
can be delineated by a set of numbers sometimes or by ordered values too.
For example, students' examination marks can be reflected by natural
numbers from 0 to 100, or by different classes A, B, and C.
