[mod.ai] Seminar - Fuzzy Relational Databases

E1AR0002@SMUVM1.BITNET (09/29/86)

Design of Similarity-Based (Fuzzy) Relational Databases
Speaker: Bill P. Buckles, University of Texas, Arlington
Location 315 SIC, Southern Methodist University
Time: 2:00PM


While the core of an expert system is its inference mechanism, a
common component is a database or other form of knowledge
representation.  The authors have developed a variation of the
relational database model in which data that is linguistic or
inherently uncertain may be represented.  The keystone concept of
this representation is the replacement of the relationship " is
equivalent to" with the relationship "is similar to".  Similarity is
defined in fuzzy set theory as an $n sup 2$ relationship over a
domain D, |D| = n such that

  i. s(x,x)=1, x member D
 ii. s(x,y)=s(y,x) x,y member D
iii. s(x,y) >= max[min(s(x,y),s(y,z))]; x, y, z member D

Beginning with a universal relation, a method is given for developing
the domain sets, similarity relationships and base relations for a
similarity-based relational database.  The universal relation itself
enumerates all domains.  The domain sets may be numeric (in which case
no further design is needed) or scalar (in which case the selection of
a comprehensive scalar set is needed).  Similarity relationship
contains $n sup 2$ values where n is the number of scalars in a domain
set.  A method is described for developing a set of consistent values
when initially given n-1 values. The base relations are derived using
fuzzy functional dependencies.  This step also requires the
identification of candidate keys.