in [5]. The present paper details the application of top-down and bottom-up algorithms
for the discovery of multivalued dependencies. The contributions of this
paper are: the denition of the hypothesis space for the discovery of multivalued
dependencies; the introduction of the data structure which allows for ecient
manipulation of the sets of multivalued dependencies; an improved procedure
for the enumeration of hypotheses; and the empirical analysis of the developed
discovery algorithms. A more detailed description of the relations between the
work on the discovery of database dependencies presented in [5] and the work
on the discovery of multivalued dependencies presented in this paper is given in
Section 5.