Database Reference
In-Depth Information
that the sensitive knowledge is hidden in the combined database which is disclosed
to the public [26]. Another thread of research in the area of association rule hiding
involves sanitization algorithms that operate on databases that are regularly updated
with new transactions instead of being static [75]. In such cases, sanitization of the
new database from scratch becomes impractical and thus should be avoided. Last,
the problem of association rule hiding has been recently considered in the context of
data streams, where unique challenges arise due to the elevated processing require-
ments of this type of data [69].
Other interesting future trends include, but are certainly not limited to, (i) the
extension of the border revision idea to cover the direct hiding of association rules,
instead of their indirect hiding through their generative itemsets, (ii) the introduc-
tion of techniques for correlation rule hiding, which is a more general field than the
one of association rules hiding, (iii) the provision and unification of more advanced
measures for the comparison of the different hiding strategies, and (iv) the inception
of spatiotemporal privacy preserving rule hiding methodologies that will prohibit
the leakage of sensitive rules related to “sensitive” spatial and/or temporal informa-
tion. The hiding of spatiotemporal patterns is currently a hot research topic since it
imposes greater challenges than the traditional knowledge hiding approaches.
Finally, future work specifically targeted in the area of exact frequent itemsets
hiding, should try to address the following research problems. First, it should in-
vestigate the possibility of further reducing the size of the constraint satisfaction
problems that are constructed by the exact hiding methodologies, while still guar-
anteeing the optimality of the hiding solution. We envision the reduction in the size
of the constraint satisfaction problems to be not in terms of how many itemsets are
controlled (i.e., each itemset should continue to be represented in the constraint sat-
isfaction problem to allow for an exact hiding solution) but in terms of the form of
the inequalities and the associated unknowns that are necessary for controlling the
status (frequent vs. infrequent) of these itemsets in the sanitized database. Second,
future work in this research area should aim at the proposal of methodologies for
the computation of the exact number of transactions that need to be appended to the
original database by the hybrid algorithm [26], in order to facilitate exact knowledge
hiding. Such methodologies will allow for a reduction in the size of the constraints
satisfaction problem that is produced by the hybrid algorithm, as well as the elimi-
nation of the post-processing phase regarding the validity of the transactions in the
generated database extension.
