Databases Reference
In-Depth Information
be presented to a user and when to stop the recursive rule generation while
querying. We refer an interested reader to [3] for details.
Granular-Computing Approach
Generally, granular computing is a broad term covering theories, methodolo-
gies, and techniques that operate with subsets, classes, and clusters (called
granules) of a universe. Granular computing concept is widely used in com-
puter science and mathematics. Recently, Zadeh [42] reviewed the concepts
of fuzzy information granulation and considered it in the context of human
reasoning and fuzzy logic. Lin [28] proposed to use the term “granular comput-
ing” to label the computational theory of information granulation. In the same
paper Lin introduces a view on DM as a “reverse” engineering of database
processing. While database processing organizes and stores data according
to the given structure, DM is aimed at discovering the structure of stored
data. Lin defines automated DM as “a process of deriving interesting (to hu-
man) properties from the underlying mathematical structure of the stored
bits and bytes” [28]. Assuming that the underlying mathematical structure
of a database relation is a set of binary relations or a granular structure, Lin
considers DM as a processing of the granules or structure-granular comput-
ing. And then if there is no additional semantics, then the binary relations
are equivalence relations and granular computing reduces to the rough set
theory [28]. However, since in the DM process the goal is to derive also the
properties of stored data, additional structures are imposed. To process these
additional semantics, Lin introduces the notion of granular computing in DM
context [27].
Yao and Yao [41] applied the granular computing approach to machine
learning tasks focusing on covering and partitioning in the process of data
mining and showed how the commonly used ID3 and PRISM algorithms can
be extended with the granular computing approach.
The Philosophy of Science Paradigm
The categorization of subjectivist and objectivist approaches [4] can be con-
sidered in the context of DM. The possibility to compare nominalistic and
realistic ontological believes gives us an opportunity to consider data that
is under analysis as descriptive facts or constitutive meanings. The analysis
of voluntaristic as opposed to deterministic assumptions about the nature of
every instance constituting the observed data directs our attitude and un-
derstanding of that data. One possibility is to view every instance and its
state as determined by the context and/or a law. Another position consists
in consideration of each instance as autonomous and independent. An episte-
mological assumption about how a criterion to validate knowledge discovered
(or a model that explains reality and allows making predictions) can be con-
structed may impact the selection of appropriate DM technique. From the
