Band Together: Untargeted Adversarial Training with Multimodal Coordination against Evasion-based Promotion Attacks

TL;DR

This paper introduces UAT-MC, a multimodal adversarial training method that aligns visual and textual perturbations to defend recommender systems against evasion-based promotion attacks, improving robustness.

Contribution

It proposes a novel gradient alignment mechanism for synchronized multimodal perturbations, addressing cross-modal mismatch in evasion attacks for recommender systems.

Findings

UAT-MC significantly enhances robustness against promotion attacks.

The method maintains recommendation performance with a favorable defense-accuracy trade-off.

Code is publicly available at https://github.com/gmXian/UAT-MC.

Abstract

Multimodal recommender systems exploit visual and textual signals to alleviate data sparsity, but this also makes them more vulnerable to evasion-based promotion attacks. Existing defenses are largely limited to single-modal settings and mainly focus on poisoning-based threats, leaving evasion-based threats underexplored. In this work, we first identify a cross-modal gradient mismatch under the multi-user promotion setting, where visual and textual perturbations are optimized in inconsistent directions due to the dominance of distinct user groups. This phenomenon dilutes the attack effectiveness and leads robust training to underestimate worst-case risks. To address this issue, we propose Untargeted Adversarial Training with Multimodal Coordination (UAT-MC). UAT-MC tackles the challenge of unknown targeted items in evasion-based attacks (as opposed to poisoning-based attacks) by treating all items as potential targets, and introduces a gradient alignment mechanism to explicitly correct this mismatch. This design ensures synchronized perturbations across modalities, thereby maximizing adversarial strength for robust training. Extensive experiments demonstrate that UAT-MC significantly improves robustness against promotion attacks while maintaining acceptable recommendation performance under the defense-accuracy trade-off. Code is available at https://github.com/gmXian/UAT-MC.