Time-aware Gradient Attack on Dynamic Network Link Prediction

In network link prediction, it is possible to hide a target from being predicted with small perturbation on structure. This observation may be exploited in many real world scenarios, for example, preserve privacy, or exploit financial security. There have been recent studies generate adversarial examples mislead deep learning models graph data. However, none of the previous work has considered dynamic nature real-world systems. this work, we present first study attack prediction (DNLP). The proposed method, namely time-aware gradient (TGA), utilizes information generated by embedding (DDNE) across different snapshots rewire few links, so as make DDNE fail predict links. We implement TGA two ways: one based traversal search, TGA-Tra; and other simplified greedy search efficiency, TGA-Gre. conduct comprehensive experiments which show outstanding performance attacking DNLP algorithms.