Home // Research // FSTM // DCS // Research Pro... // Give control back to users: personalised privacy-preserving data aggregation from heterogeneous social graphs - resubmission

Give control back to users: personalised privacy-preserving data aggregation from heterogeneous social graphs - resubmission

Funding: Fonds National de la Recherche > CORE
Start Date: March 1, 2021
End Date: Feb. 29, 2024


Heterogeneous social graphs (HSG) have been widely used to analyse social network data to support decision making. Compared to simple social graphs which only model the relations between users, HSGs capture the heterogeneity nature of social networks in terms of data subjects and relations between them. The richer information encoded in HSGs leads to overwhelming better results than those on simple social graphs. In the meantime, it also imposes more risk of a privacy breach. Due to the potential economic and reputation loss, social network operators only publish a limited amount of HSG data for researchers and third-party data analysts.In this project, we address an alternative decentralised solution for data analysts to collect data of HSGs directly from volunteers while guaranteeing volunteers’ privacy. Specifically, users privately calculate and share data about their local views of HSGs. Data analysts aggregate these responses into the information of interest. To the best of our knowledge, no works in the literature exist to achieve this goal. Moreover, we will take into account the fact that in real-life scenarios, users may have different privacy requirements, e.g., due to various trust to data collectors. We design methods for users to perturb their local data according to their own personalised privacy requirements. In this manner, we manage to give control back to users over their data by determining the level of privacy protection. In addition to precise privacy preservation, our methods can also ensure better utility for the aggregated data when only a small number of users require high-level protection.To achieve our purpose, we will first extend the notion of local differential privacy to quantify users' personalised privacy requirements over different types of sensitive information, i.e., vertices and edges. Once the privacy properties have been defined, we will design corresponding privacy-preserving methods for two widely studied data aggregation tasks: query answering and graph synthesis. Query answering is used to aggregate statistics of some structural properties of HSGs while graph synthesis allows data analysts to conduct flexible analysis on synthetic HSGs with similar properties to the original graphs. Last but not least, we will develop a comprehensive evaluation framework to evaluate the effectiveness of our methods and define new measures to quantitatively assess the utility of the aggregated data.