Exploiting contextual information in attacking set-generalized\ud transactions

摘要

Transactions are records that contain a set of items about individuals. For example, items browsed by a\udcustomer when shopping online form a transaction. Today, many activities are carried out on the Internet,\udresulting in a large amount of transaction data being collected. Such data are often shared and analyzed\udto improve business and services, but they also contain private information about individuals that must\udbe protected. Techniques have been proposed to sanitize transaction data before their release, and setbased\udgeneralization is one such method. In this paper, we study how well set-based generalization can\udprotect transactions. We propose methods to attack set-generalized transactions by exploiting contextual\udinformation that is available within the released data. Our results show that set-based generalization may\udnot provide adequate protection for transactions, and up to 70% of the items added into the transactions\udduring generalization to obfuscate original data can be detected by our methods with a precision over 80%.