ShareSafe

ShareSafe is a uniform and open-source secure graph data sharing/publishing system. In ShareSafe, we systematically study, implement, and evaluate graph data anonymization algorithms, data utility metrics, modern Structure-based De-Anonymization (SDA) attacks, and theoretical Structure-based De-Anonymization Quantification techniques. ShareSafe is a comprehensive system that enables data owners and researchers to anonymize data by state-of-the-art anonymization techniques, measure the data’s utility, and evaluate the data’s vulnerability against modern attacks.

Anonymization Module (AM)

This module can anonymize raw graph data and generate anonymized data. In this module, we implement 11 state-of-the-art graph data anonymization schemes, including Edge Editing based algorithms, k-anonymity based algorithms and its variants, aggregation/class/cluster based algorithms, differential privacy based algorithms, and the random walk based algorithm.

Utility Module (UM)

This module can evaluate raw/anonymized data’s utility with respect to the 12 graph utility metrics and 7 application utility metrics. With the UM, we can determine whether the data to be published/shared (e.g., the anonymized data) satisfies required utility requirements. We can also evaluate how an anonymization algorithm preserves data utility.

De-Anonymization Module (DM)

This module offers 15 structural based de-anonymization algorithms (SDA) (all the existing SDA algorithms, to the best of our knowledge). In this module, the security of data to be published/shared can be evaluated with real-world powerful SDA attacks. More importantly, the effectiveness of an anonymization algorithm can be examined by this module, i.e., whether the anonymized data of an anonymization algorithm is resistant to modern SDA attacks.

De-Anonymizability Quantification Module (DQM)

Based on our structural importance-aware adaptive quantification, we implement a Relative De-Anonymizability Quantification Meter (RDQM). Given a graph dataset, RDQM can quantitatively demonstrate how many and which users are de-anonymizable according to the dataset’s structural information. Therefore, this module can quantitatively examine the effectiveness of an anonymization scheme. Furthermore, the quantification results can serve as auxiliary information that is useful for future anonymization and DA algorithms design.