SP-GCRL: AI Framework for Influence Maximization on Incomplete Social Graphs

Researchers have introduced SP-GCRL, a social-propagation-aware graph contrastive reinforcement learning framework designed to address influence maximization (IM) on incomplete and noisy social graphs with non-stationary diffusion dynamics. The framework learns end-to-end seed selection under partial observability. It incorporates a nonlinear diffusion function to model reinforcement, diminishing effects, and probability drift from repeated exposure. Dual structural views and contrastive learning produce node representations robust to missing edges and weak ties. A GAT-based regression surrogate replaces expensive strategy metrics for efficiency. A Double Deep Q-Network (DDQN) learns the seed selection policy. Experiments on real-world networks show significant gains over heuristic and learning-based baselines. The paper is available on arXiv under ID 2605.12513.