A GEOMETRICAL FUZZY PARTITIONS APPROACH TO FUZZY QUERY AND FUZZY DATABASE
RETRIE VAL.
E. G. Mtalo and E. Derenyi
Department of Surveying Engineering
University of New Brunswick
Fredericton, N.B. CANADA
COMMISSION III
ABSTRACT
Real world knowledge is usually vague and ambiguous and human beings generally think and communicate in fuzzy non-precise
terms. The fuzzy sets theory introduced by Zadeh in 1965 forms the mathematical and practical basis for the representation and
manipulation of such fuzzy information.
Conventional databases, however, must contain precisely defined facts or data because database query languages such as the SQL
impose strict formats for data entry and query. Conventional query languages do not permit ambiguous or non-precise queries .
This paper presents a new method for the representation of fuzzy numerical quantities in a way favorable to the storage and retrieval of
fuzzy values or vague expressions in a database.
Key Words: Fuzzy Sets, Fuzzy Retrieval, Geometric Partitions, GIS, Database.
1. INTRODUCTION.
Conventional databases must contain precisely defined facts and
numeric values because database query languages such as the
SQL impose strict formats for data entry and query.
Conventional query languages do not permit ambiguous or
non-precise queries. Real world data and knowledge is usually
vague and ambiguous and human beings generally think
vaguely and communicate in fuzzy non-precise terms (Zadeh et
al, 1975; Zadeh, 1989). However vague or fuzzy real world
knowledge does not lend itself to easy manipulation by
conventional database management systems.
The fuzzy sets theory introduced by Zadeh in 1965 (Zadeh et al.
1975; Kandel, 1986) forms the mathematical and practical basis
for the representation and manipulation of fuzzy information.
1.1 Representation of Vague Information by Fuzzy
Sets.
Consider the statement that the distance from the Earth to the
Sun is "very great", or the statement "100 is much greater than
5". Using Zadeh's fuzzy sets theory, the terms "great" and
"greater" may be regarded as fuzzy sets and can therefore be
defined in terms of fuzzy membership functions. Once these
fuzzy sets have been defined the modifiers, "very" and "much",
can be applied to transform them into the corresponding fuzzy
sets very great and much greater respectively(see Zadeh et al.
1975; Kaufmann and Gupta, 1988; Shmucker, 1984; Kandel,
1986; Zadeh, 1989).
Typically the fuzzy set great will be represented by a fuzzy
membership function Mgrea(X): X-2[0,1] and greater can be
represented by the membership function
Hgreater than(X): X->{0,1]. Generally MAX) denotes the
membership function of the elements of the universe X in the
fuzzy set A such that the elements, x, take value in the
universe X. The membership values, HA(X) on the other hand
take value in the evaluation space of the fuzzy set, generally
considered to be the continuous interval [0,1] (Zadeh et al,
1975; Dubois and Prade, 1980; Kandel, 1986). The
membership value Ha(x), is a real number between O and 1
inclusive, expressing the strength of the membership of an
element, x, of the universe X in a the fuzzy set A.
Practical specification of a membership function involves the
assignment of the parameters of some standard membership
function, such as the S-function and x-function (Dubois and
Prade 1980; Kandel 1986; Klir and Folger 1988). Important
characteristic points of standard membership functions include;
the cut-out points (points with zero membership value), the
peak point (where the membership function attains the
176
maximum value of 1) and the turnover points (points where the
membership function has the value of 0.5). Dubois and Prade
(1988, 1990) suggest that for most applications simpler piece-
wise linear membership functions such as triangular functions
and trapezoidal functions provide satisfactory results.
The method proposed in this paper differs from the standard
approach in that, fuzzy expressions such as "greater than five"
are characterized by geometric partitions induced by them in the
real numbers domain. The paper begins by defining the concept
of fuzzy geometric partitions and then goes on to show how it
can be applied to the problem of storing and querying fuzzy
database objects. For the purposes of fuzzy database retrieval,
the universe of discourse is some search space X containing
crisp and fuzzy objects. To facilitate database retrieval each
fuzzy query and fuzzy database object can be associated with a:
fuzzy partition defined over the search space. The condition for
a successful search is obtained when the partition induced by
the query object contains the partition generated by the fuzzy
database object.
The paper also looks at the potential use of the fuzzy
geometrical partitions method in the construction of membership
functions, which may then be used to represent fuzzy numerical
sets in the usual way(Zadeh ,1975, 1989).
2. THE CONCEPT OF FUZZY GEOMETRIC
PARTITIONS.
The concept of fuzzy partitions is not new, what is new is,
however, the manner in which this concept is used to
characterize and represent fuzzy valued expressions or fuzzy
numbers. An earlier use of the term radial partition to
characterize fuzzy sets can be found in Kaufmann (1975).
Kaufmann shows that a fuzzy set, M, induced by the binary
relation y >> x , for y = kx, k > 1, in the two dimensional real
space constitutes a radial partition of the real numbers space.
In Dowsing et al.(1986) the concept of the diagonal set of the
universe of discourse, is used to define and characterize the
equality operator. This research extends and generalizes the
concept of the diagonal set generated by the equality operator,
defined in Dowsing et al.(1986), and uses it to define general
fuzzy comparison operators or fuzzy binary relations, R(x,y),
in terms of the radial sets or partitions induced by them in a two
dimensional real space X.
Intuitively the operators "equal", "greater", and "less" define
basic geometric partitions of the two dimensional space (Figure
1). The partition generated by the equality operator, =, is called
the diagonal set (Dowsing et al, 1986) or diagonal partition. By
extension the partition corresponding to the operator greater (>)
is called the upper diagonal partition, while the partition of the