How to Prevent Private Data from being Disclosed to a Malicious Attacker - Data Mining: Foundations and Practice - page 502

How to Prevent Private Data
from being Disclosed to a Malicious Attacker

Justin Zhan 1 , LiWu Chang 2 , and Stan Matwin 3

1

Carnegie Mellon University
[e-mail]

2

Naval Research Laboratory
[e-mail]

3

University of Ottawa
[e-mail]

1 Introduction

In this paper, 1 we address the following problem: multiple parties are coop-
erating on a data-rich task. Each of the parties owns data pertinent to the
aspect of the task addressed by this party. More specifically, the data consists
of instances, each party owns her instances but all parties have the same at-
tributes. The overall performance, or even solvability, of this task depends on
the ability of performing data mining using all the instances of all the parties.
The parties, however, may be unwilling to release their instances to other
parties, due to privacy or confidentiality of the data. How can we structure
information sharing between the parties so that the data will be shared for
the purpose of data mining, while at the same time specific instance values
will be kept confidential by the parties to whom they belong? This is the
task addressed in this paper. In the privacy-oriented data mining this task
is known as data mining with horizontally partitioned data (also known as
homogeneous collaboration [15]). Examples of such tasks abound in business,
homeland security, coalition building, medical research, etc.
The following scenarios illustrate situations in which this type of collabora-
tion is interesting: (i) Multiple competing supermarkets, each having an extra
large set of data records of its customers' buying behaviors, want to conduct
data mining on their joint data set for mutual benefit. Since these companies
are competitors in the market, they do not want to disclose their customers'
information to each other, but they know the results obtained from this col-
laboration could bring them an advantage over other competitors. (ii) Success
of homeland security aiming to counter terrorism depends on combination of
strength across different mission areas, effective international collaboration

1 The preliminary version of this paper has been published [32].
J. Zhan et al.: How to Prevent Private Data from being Disclosed to a Malicious Attacker ,