Information Technology Reference
In-Depth Information
Fig. 6.
A problematic data sample for the FD
Title
→
National number, Last
name, First name, Birth date, Contact
, given the valid data samples of Fig. 4
dependency must be satisfied by the existing set of tuples associated with the
considered entity type.
Alternatively, the participants can also take advantage of the potential con-
straints and dependencies to arbitrate them, i.e. to validate or discard them. The
advantages of this approach are that the participants do not have to imagine all
the possible constraints and dependencies for each entity type, and that we di-
rectly know that each candidate constraint or dependency is currently potential
for the given entity type. One can suspect that validating or discarding a con-
straint or a dependency may impact on the constraints or dependencies of other
types. Such a correlation actually exists between functional dependencies and
unique constraints. Indeed, discarding a potential functional dependency may
change the potential unique constraints, whereas validating a unique constraint
automatically validates its underlying functional dependency and discards oth-
ers. When these cases occur, the relevant sets must therefore be updated.
Besides, it obviously appears that the number of suggested constraints and
dependencies can eventually become very high. It is therefore crucial to organ-
ise these suggestions in an accessible fashion, so that the end-users do not feel
overwhelmed. Besides, this underlines the importance of the analyst to guide
the end-users through this collaborative process, by assessing the relevance of
these suggestions. This observation is especially true regarding the elicitation
of the functional dependencies, since the number of suggestions can increase
dramatically.
We therefore propose to filter the potential functional dependencies in or-
der to limit the number of relevant suggestions, while privileging the “stronger”
functional dependencies (i.e. the dependencies with smaller left-hand side and
larger right-hand side, as previously explained). For this purpose, we therefore
propose to “hide” dependencies that can be obtained from other potential de-
pendencies using
Armstrong's axioms
, which are a set of inference rules used to
infer all the functional dependencies on a relation [19]. Hiding these functional
