Strategies for Knowledge Based Attack Detection in Graphical Data Anonymization

Abstract: Recent years have seen a rapid growth in web applications developed, which gave rise to the increasing need for private data to be published. Most of the social network data necessitates the data to be available for easy access and conversion of data to graph structure to re-identify sensitive labels of individuals became an impeccable issue. Privacy protection scheme that not only prevents the disclosure of identify of users but also the disclosure of selected features in users’ profile. Existing KDLD (k-degree l-diversity) model for ensuring anonymity of data, have some restrained but severe privacy problems. Firstly, an attacker can discover the values of sensitive attributes when there is little diversity in those sensitive attributes leading to homogenity attack. Secondly, an adversary could often use background knowledge to discover sensitive information, causing a background knowledge attack. Thirdly, when the sensitive attribute values in an equivalence class are distinct but semantically similar, an adversary can learn important information, rooting to the problem of similarity attack. This paper considers the problem of detecting these three attacks in a k-degree-l-diversity (KDLD) graph based data and to address the same on graphical data.

Keywords: Privacy, Graphical Data, Re-identification, Attack, Data Security, Anonymization