Secure Incremental Maintenance of Distributed Association Rules

Speaker: WONG Wai Kit

Abstract

With the rapid growth of Internet, distributed mining of association rules helps to find associated itemsets from transactions produced and stored in a distributed network. In order to discover more accurate and useful knowledge, branches of the same companies or even different companies may want to cooperate in the data mining process without disclosing any (original or derived) information of their own data to the other parties. Research results have been developed on (i) incrementally maintaining the discovered association rules, and (ii) computing the distributed association rules while preserving privacy. However, no study has been conducted on the problem of the maintenance of the discovered rules with privacy preservation when new sites are allowed to join the old sites. Re-computation of the rules distributively on the new and old sites together cannot be accepted because it essentially wastes all the rules found. We propose two maintenance algorithms SIMDAR and MSIMDAR for this problem that provide two different security levels of data privacy. Experimental results showed that MSIMDAR can significantly reduce the workload at the old sites by up to 70%.