Information Technology Reference
In-Depth Information
Definition 7.10
A concept level is a mapping sequence from inferior concepts to
superior concepts. These mapping sequences are organized in tree to generate a
concept level tree.
0
Definition 7.11
corresponds to a concept
level forest F ={F
1
,
…
, F
i
,
…
, F
ϖ
}, where F
i
is a set of concept level tree
corresponding to the ith attribute Ai, 1
≤
i
≤
ϖ
. F
i
= {T
1
,
…
, T
j
,
…
,T
τ
}, where T
j
is
the jth concept level tree of attribute A
i
, 1
≤
j
≤
τ
.
A primitive training example set E
Definition 7.12
When there is no concept level tree in attribute A
i
, F
i
= NULL.
Definition 7.13
A concept level database D is used to save history concept level
trees.
In all attributes of primitive training set, many attributes have their own fixed
concept level. For example, if a commodity is made in Beijing, we can say it is
made in China. Furthermore, we also can say it is made in Asia. This is a concept
level of {Product Places : Beijing, China, Asia}.
Concept level is used to represent requested background knowledge to
control procedure of generalization or specialization. Through organizing
concepts of different level into a tree classification, concept space can be
represented as partial order from specialization to generalization. The most
general concept is description with empty meaning, which can be represented as
“ANY”; the most specific concept is leaf in tree classification. Making use of
concept level, we can represent found rules as form simpler, easier more special
and more logical.
7.8.3
Algorithms
1. Classification guided attribute oriented inductive algorithm CGAOI
Classification guided attribute oriented inductive algorithm CGAOI can
generalize original relation into appointed concept level. We proposed
Classification guided attribute oriented inductive algorithm CGAOI. This
approach is a supervised approach. As guide of classification learning task, it
preprocesses primitive training instances, and softly outputs generalization
relation of appointed level.
In the guide of classification feature of primitive training instances, the
algorithm realizes basic operation oriented by attribute induction, such as
attribute eliminating, concept refining, attribute threshold controlling and
