Adaptive Graph Neural Networks with Curriculum Learning for Robust Social Network Analysis

Graph Neural Networks (GNNs) are critical for social network analysis but vul- nerable to adversarial edge perturbations, degrading bot detection on platforms like Twitter. We propose a GNN with an adaptive curriculum learning (CL) scheduler that dynamically adjusts subgraph complexity based on training loss, enhancing robustness. Sparse subgraphs are sampled early, scaling to denser graphs as learning stabilizes, yielding a 5% improvement in accuracy and robust- ness over baseline GCNs on a synthetic Twitter dataset. This work advances trustworthy AI for social network security, offering a scalable solution for global research communities with applications to bot detection and beyond.