Online Social Network De-anonymization via Conditional Generative Adversarial Network Model
Primary Investigator:
Feng Li
Tianchong Gao; Feng Li
Abstract
Nowadays, Online Social Network (OSN) information leakage raises more and more concerns. Studying the problem from both attack and defense perspectives provides valuable insights to researchers on OSN privacy preservation. Several anonymization mechanisms based on K-anonymity and differential privacy are newly proposed, yet the innovation of de-anonymization mechanisms struggles with the complexity of graph structures and the difficulty of finding the global optimal mapping strategy.
This project introduces machine learning methods to analyze graph structures. While existing de-anonymization mechanisms collect information by mapping users from adversary’s background knowledge to published data, this project directly generates a graph containing the link information and attribute information of targets. In particular, the property of the Generative Adversarial Network (GAN) ensures that the generated graph is undistinguishable with the published graph. The adversaries’ background knowledge is embedded as conditional information into the GAN. We also adopt the specially designed graph auto-encoder and graph neural network to extract graph features. The evaluation on real-world OSN datasets proves that the new de-anonymization scheme can generate new graphs similar to the original OSN, de-anonymize the edge information with high accuracy, and enhance the existing user de-anonymization schemes.