SI2-SSE: ShareSafe: A Framework for Researchers and Data Owners to Help Facilitate Secure Graph Data Sharing

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

There is a critical need for information/data sharing to solve some of our most significant academic and societal problems. These data are increasing in size and are becoming much more complex; in many cases, they can be considered structured. An example of structured data is data describing disease propagation in a specific population. Widespread sharing of data can, among many other things, help corporations increase their revenues, help reduce the spread of communicable diseases, accelerate the cure of some of the most significant diseases, and enable reproducible experiments amongst researchers. Although there is little disagreement that sharing data has tremendous benefits, it is still not as widespread as it should be. This is, in part, due to privacy concerns with sharing datasets. This project will develop an open source system (ShareSafe) that allows data owners to evaluate the security (such as resistance to de-anonymization attacks) and utility of their anonymized datasets before release, which will help facilitate the data sharing process. The overarching goals of this project are to develop a software framework, ShareSafe, that (1) helps structured data owners (e.g., social network researchers, epidemiologists) evaluate the security (against modern de-anonymization attacks) and utility of their datasets when using simple and state-of-the-art anonymization techniques; and (2) to provide structured data security/privacy researchers a uniform platform to comprehensively study, evaluate, and compare existing/newly developed techniques for structured data utility and privacy. ShareSafe is a comprehensive, user-friendly framework with the following capabilities: ShareSafe will enable data owners to: (1) anonymize their datasets with all of the state-of-the art anonymization techniques; (2) measure the utility of anonymized datasets using state-of-the-art utility measurement techniques; (3) evaluate the practical security of their datasets by subjecting them to state-of-the-art de-anonymization attacks; and (4) evaluate the theoretical security of their datasets by subjecting them to state-of-the-art de-anonymization quantification (de-anonymizability analysis) techniques. Understanding the results from (2)-(4) allows data owners to determine which anonymization algorithm suits their needs when sharing datasets. Finally, the aforementioned techniques will be implemented in a uniform manner as open source software, allowing graph data security/privacy researchers the ability to comprehensively study, evaluate, and compare existing/newly developed techniques for graph data utility and privacy.