Feature-Based Graph Backdoor Attack in the Node Classification Task

Graph neural networks (GNNs) have shown significant performance in various practical applications due to their strong learning capabilities. Backdoor attacks are a type of attack that can produce hidden on machine models. GNNs take backdoor datasets as input an adversary-specified output poisoned data but perform normally clean data, which grave implications for applications. under-researched the graph domain, and almost existing focus graph-level classification task. To close this gap, we propose novel uses node features triggers does not need knowledge parameters. In experiments, find feature destroy spaces original datasets, resulting inability identify well. An adaptive method is proposed improve model by adjusting structure. We conducted extensive experiments validate effectiveness our three benchmark datasets.