Database Reference
In-Depth Information
Chapter 20
Conclusions
The serious privacy concerns that are raised due to the sharing of large transactional
databases with untrusted third parties for association rule mining purposes, soon
brought into existence the area of association rule hiding, a very popular subarea
of privacy preserving data mining. Association rule hiding focuses on the privacy
implications originating from the application of association rule mining to shared
databases and aims to provide sophisticated techniques that effectively block access
to sensitive association rules that would otherwise be revealed when mining the data.
The research in this area has progressed mainly along three principal directions:
(i) heuristic-based approaches, (ii) border-based approaches, and (iii) exact hiding
approaches. Taking into consideration the rich work proposed so far, in this topic
we tried to collect the most significant research findings since 1999, when this area
was brought into existence, and effectively cover each principal line of research.
The detail of our presentation was intentionally finer on exact hiding approaches,
since they comprise the most recent direction, offering increased quality guarantees
in terms of distortion and side-effects introduced by the hiding process.
The first part of the topic serves as an introduction to the area of association
rule hiding by motivating the problem at hand, discussing its predominant research
challenges and variations, and providing the necessary background for its proper
understanding. The distinctive characteristics of the proposed methods led us to
propose a taxonomy to partition them along four orthogonal dimensions based on
the employed hiding strategy, the data modification strategy, the number of rules that
are concurrently hidden, and the nature of the algorithm. By using this partitioning,
we devoted each subsequent part of the topic to a specific class of approaches, hence
presenting heuristic-based approaches in Part II, border-based approaches in Part III,
and exact hiding approaches in Part IV of the topic.
The majority of the proposed methodologies for association rule hiding are of
heuristic nature in order to effectively tackle the combinatorial nature of the prob-
lem. The basic property of heuristic algorithms that makes them attractive is their
computational and memory efficiency, which allows them to scale well even to very
large datasets. A wide variety of heuristics have been proposed over the years. In the
second part of the topic, we partitioned the heuristic methodologies along two main
Search WWH ::




Custom Search