Practical Cross-System Shilling Attacks with Limited Access to Data

In shilling attacks, an adversarial party injects a few fake user profiles into Recommender System (RS) so that the target item can be promoted or demoted. Although much effort has been devoted to developing attack methods, we find existing approaches are still far from practical. this paper, analyze properties practical method should have and propose new concept of Cross-system Attack. With idea Attack, design Practical Shilling Attack (PC-Attack) framework requires little information about victim RS model data for conducting attacks. PC-Attack is trained capture graph topology knowledge public in self-supervised manner. Then, it fine-tuned on small portion easy access construct profiles. Extensive experiments demonstrated superiority over state-of-the-art baselines. Our implementation available at https://github.com/KDEGroup/PC-Attack.