RoBCtrl: Attacking GNN-Based Social Bot Detectors via Reinforced Manipulation of Bots Control Interaction

TL;DR

RoBCtrl presents a novel adversarial framework that links diffusion-based bot feature generation (DiffBot) with multi-agent reinforcement learning to attack GNN-based social bot detectors under black-box constraints. By modeling bot heterogeneity and coordinating evolving bot strategies through a structural-entropy–driven state abstraction, it demonstrates substantial degradation of detector performance on large- and normal-scale graphs while preserving many global graph properties. The work highlights a critical vulnerability in current GNN-based detectors and shows that state abstraction can significantly boost attack efficiency. These findings underscore the need for robust, group-aware defenses and responsibly managed deployment of bot-detection systems in real social networks.

Abstract

Social networks have become a crucial source of real-time information for individuals. The influence of social bots within these platforms has garnered considerable attention from researchers, leading to the development of numerous detection technologies. However, the vulnerability and robustness of these detection methods is still underexplored. Existing Graph Neural Network (GNN)-based methods cannot be directly applied due to the issues of limited control over social agents, the black-box nature of bot detectors, and the heterogeneity of bots. To address these challenges, this paper proposes the first adversarial multi-agent Reinforcement learning framework for social Bot control attacks (RoBCtrl) targeting GNN-based social bot detectors. Specifically, we use a diffusion model to generate high-fidelity bot accounts by reconstructing existing account data with minor modifications, thereby evading detection on social platforms. To the best of our knowledge, this is the first application of diffusion models to mimic the behavior of evolving social bots effectively. We then employ a Multi-Agent Reinforcement Learning (MARL) method to simulate bots adversarial behavior. We categorize social accounts based on their influence and budget. Different agents are then employed to control bot accounts across various categories, optimizing the attachment strategy through reinforcement learning. Additionally, a hierarchical state abstraction based on structural entropy is designed to accelerate the reinforcement learning. Extensive experiments on social bot detection datasets demonstrate that our framework can effectively undermine the performance of GNN-based detectors.

Figures (6)

• Figure 1: The proposed RoBCtrl framework.
• Figure 2: Proposed RoBCtrl framework optimized by hierarchical state abstraction.
• Figure 3: The proportion of source node and target node types before and after the RoBCtrl's attack.
• Figure 4: Distributions of node degrees before and after attack. (Note: The white line always exists inside the black line.)
• Figure 5: Accuracy and attack time results after attacked by RoBCtrl and State Abstraction in normal-scale social bot graph datasets.

Theorems & Definitions (1)

• Definition 4.1: Markov Game