A New Approach for Enhancing Security in Anonymizing Transaction Data with Set Values
Jiyeon Lee, Donghyun Kim, Soonseok Kim · Asia-pacific Journal of Convergent Research Interchange · 2024
In relational databases, there exist instances where a cell does not simply contain a singular character or number but rather a series of values, constituting a set of items.Such data configurations are generally referred to as transaction data.For instance, this can include the assortment of products specific customers purchase at a supermarket.This paper aims to explore methods for anonymizing transaction data derived from structured microdata, thereby safeguarding the personal information of the data subjects.While effective in specific contexts, the existing algorithms exhibit limitations, particularly when handling items of a sensitive nature (e.g., supercars, VIP room reservations) that a particular customer has purchased, which may necessitate additional protection measures for privacy preservation.Depending on the application, further protective measures might be required, such as: firstly, concealing the fact that a particular customer has purchased a specific set of sensitive items, denoted as α(={α1, α2, α3, ..., αn}), from potential attackers; secondly, ensuring that the items within a transaction purchased by a specific customer include a minimum of β diverse items; and, thirdly, situations where both of the conditions as mentioned earlier are simultaneously requisite.This research proposes a novel algorithm that enhances the security of the previously suggested algorithms by incorporating three constraints that address the necessary protection measures.The proposed algorithm, building upon the k-anonymity model, not only maintains the 1/k chance of re-identification but also introduces an additional layer of security by considering α as the set of specific sensitive items comprising n elements designated by a particular customer and β as the number of items purchased, thereby offering an augmented safety margin of 1/k+n+1/β.