A Theoretically-Sound Accuracy/Privacy-Constrained Framework for Computing Privacy Preserving Data Cubes in OLAP Environments

The problem of ensuring the privacy and security of OLAP data cubes (Gray et al., 1997) arises in several fields ranging from advanced Data Warehousing (DW) and Business Intelligence (BI) systems to sophisticated Data Mining (DM) tools. In DW and BI systems, decision making analysts aim at avoiding that malicious users access perceptive ranges of multidimensional data in order to infer sensitive knowledge, or attack corporate data cubes via violating user rules, grants and revokes. In DM tools, domain experts aim at avoiding that malicious users infer critical-for-thetask knowledge from authoritative DM results such as frequent item sets, patterns and regularities, clusters, and discovered association rules. In more detail, the former application scenario (i.e., DW and BI systems) deals with both the privacy preservation and the security of data cubes, whereas the latter one (i.e., DM tools) deals with privacy preserving OLAP issues solely. With respect to security issues, although security aspects of information systems include a plethora of topics ranging from cryptography to access control and secure digital signature, in our work we particularly focus on access control techniques for data cubes, and remand the reader to the active literature for the other orthogonal matters. Specifically, privacy preservation of data cubes refers to the problem of ensuring the privacy of data cube cells (and, in turn, that of queries defined over collections of data cube cells), i.e. hiding sensitive information and knowledge during data management activities, according to the general guidelines drawn by Sweeney in her seminar paper (Sweeney, 2002), whereas access control issues refer to the problem of ensuring the security of data cube cells, i.e. restricting the access of unauthorized users to specific sub-domains of the target data cube, according to well-known concepts studied and assessed in the context of DBMS security. Nonetheless, it is quite straightforward foreseeing that these two even distinct aspects should be meaningfully integrated in order to ensure both the privacy and security of complex data cubes, i.e. data cubes built on top of complex data/knowledge bases. During last years, these topics have became of great interest for the Data Warehousing and Databases research communities, due to their exciting theoretical challenges as well as their relevance and practical impact in modern real-life OLAP systems and applications. On a more conceptual plane, theoretical aspects are mainly devoted to study how probability and statistics schemes as well as rule-based models can be applied in order to efficiently solve the above-introduced problems. On a more practical plane, researchers and practitioners aim at integrating convenient privacy preserving and security solutions within the core layers of commercial OLAP server platforms. Basically, to tackle deriving privacy preservation challenges in OLAP, researchers have proposed models and algorithms that can be roughly classified within two main classes: restriction-based techniques, and data perturbation techniques. First ones propose limiting the number of query kinds that can be posed against the target OLAP server. Second ones propose perturbing data cells by means of random noise at various levels, ranging from schemas to queries. On the other hand, access control solutions in OLAP are mainly inspired by the wide literature developed in the context of controlling accesses to DBMS, and try to adapt such schemes in order to control accesses to OLAP systems.

Download Full-text