RABot: Reinforcement-Guided Graph Augmentation for Imbalanced and Noisy Social Bot Detection

A research team led by Longlong Zhang, Xi Wang, and Yang Liu has introduced RABot, a novel framework designed to overcome persistent challenges in social bot detection. The system addresses two critical problems that plague current graph neural network (GNN) approaches: severe class imbalance caused by the high cost of generating labeled bot data, and topological noise introduced by sophisticated bots that mimic human behavior and create deceptive connections. RABot's unified approach combines graph augmentation techniques with reinforcement learning to create more robust detection systems that can operate effectively in real-world, noisy environments where traditional methods struggle.

Technically, RABot employs a neighborhood-aware oversampling strategy that performs linear interpolation of minority-class embeddings within local subgraphs, stabilizing decision boundaries in low-resource scenarios. Simultaneously, a reinforcement-learning-driven edge-filtering module combines similarity-based edge features with adaptive threshold optimization to remove spurious interactions during message passing, resulting in cleaner network topologies. The framework demonstrated superior performance across three real-world benchmarks and four different GNN backbones, showing consistent improvements over state-of-the-art baselines. Since RABot's augmentation and filtering modules are architecture-agnostic, they can be seamlessly integrated into existing GNN pipelines to boost detection performance with minimal computational overhead, making it practical for large-scale social platform deployment.