Adversarial Privacy-Preserving Graph Embedding Against Inference Attack

Recently, the surge in popularity of Internet Things (IoT), mobile devices, social media, etc., has opened up a large source for graph data. Graph embedding been proved extremely useful to learn low-dimensional feature representations from graph-structured These can be used variety prediction tasks node classification link prediction. However, existing methods do not consider users' privacy prevent inference attacks. That is, adversaries infer sensitive information by analyzing learned algorithms. In this article, we propose adversarial (APGE), training framework that integrates disentangling and purging mechanisms remove private representations. The proposed method preserves structural utility attributes while concealing Extensive experiments on real-world data sets demonstrate superior performance APGE compared state-of-the-arts. Our code found at https://github.com/KaiyangLi1992/Privacy-Preserving-Social-Network-Embedding.