Information Technology Reference
In-Depth Information
inductive assertion relating these descriptors. (2) related partially, there are many
useless or redundant descriptors. Some of them are related. Now learning task is
to choose the most related descriptors, and construct rational inductive assertion
based on these descriptors. (3) related indirectly, observation statement does not
include descriptors directly related to problems. However, in initial description,
some descriptors can be used to generate related descriptors. The task of learning
is to generate these related descriptors directly, and get inductive assertion.
Descriptor labeling is background knowledge set related to descriptors and
learning problems. It includes:
Type description of definition domain and descriptors;
Operator description related to descriptors;
Constraints and relation description among descriptors;
Significance and vary law of descriptors representing number in problems;
Features of descriptors applicable to objects;
Class including given descriptor, i.e. parent node of descriptor;
Synonym that could replace the descriptor;
Definition of descriptors;
Typical example of given descriptors of objects.
Definition domain of descriptor is a set of values that descriptor can have. For
example, human's body temperature is between 34
o
C and 44
o
Cthen the value of
descriptor “body temperature” can only be in this range. Descriptor type is
decided by relation of elements in definition domain of descriptor. According to
the structure of definition domain of descriptor, there are three basic types:
(1) Nominal descriptor. Definition domain of this descriptor is composed of
independent symbol or name. That is, values in value set have not structural
relation, such as fruit, people's name.
(2) Linear descriptor. Elements in value set of this kind of descriptors are in
accordance with a totally ordered set. For example, funds, temperature, weight,
product are all linear descriptors. Variables representing ordinal number, interval,
ratio and absolute calibration are specific instances of linear descriptor. Functions
which map a set to a totally ordered set are linear descriptors too.
(3) Structural descriptor. Its value set is a tree forming graphical structure,
reflecting generative levels among values. In such structure, parent node
represents more general concept than child node. For example, in value set
“place name”, “China” is parent of “Beijing”, “Shanghai”, “Jiangsu”,
“Guangdong”, etc. Definition domain of structural descriptor is defined by a
group of reasoning rules explained by background knowledge of problems.
Structural descriptor can be further subdivided into ordered and unordered
