A Differentially Private Approach for Querying RDF Data of Social Networks

Abstract

As the amount of collected social network information in RDF format grows, the development of solutions for the privacy of individuals, their attributes and relationships with others becomes an important subject of study. However, data privacy solutions are not well suitable for this specific type of data, mainly because they usually do not consider relationships between individuals, which are crucial to semantic data and social networks. Differential privacy is one of the most suitable techniques for statistical queries and, although it has been extensively studied in many papers, there is still much research to be done in this context. This paper presents two main contributions for privacy preserving statistic queries containing sensitive information about relationships between individuals. The first one is a complete approach to applying ϵ-differential privacy for RDF data and the second one presents an index-like data structure to efficiently compute parameters for the differential privacy mechanism: the query's actual value and data sensitivity for the given query. We conclude by evaluating our contributions over three real social network datasets presenting utility analysis for different values of ϵ. We also show the performance benefit of our index-like data structure for sensitivity calculation.