Problem
Research questions and friction points this paper is trying to address.
Influence Maximization
Incomplete Social Graphs
Non-stationary Diffusion
Partial Observability
Noisy Graphs
SP-GCRL: Influence Maximization on Incomplete Social Graphs
🤖 AI Summary
This work addresses the challenge of influence maximization in real-world social networks, where incomplete graph structures, noisy observations, and non-stationary diffusion dynamics hinder effective seed selection. To tackle this problem, the authors propose SP-GCRL, a novel framework that integrates a propagation-aware nonlinear diffusion mechanism, dual-view graph contrastive learning, and a double deep Q-network (DDQN) to enable end-to-end seed selection under partial observability. By leveraging a graph attention network (GAT) to construct an efficient regression surrogate, SP-GCRL bypasses costly policy evaluations, substantially enhancing both robustness and computational efficiency. Extensive experiments demonstrate that SP-GCRL consistently outperforms state-of-the-art heuristic and learning-based baselines across multiple real-world networks, maintaining superior performance and strong scalability under varying budgets and topological conditions.
📝 Abstract
Influence maximization (IM) in real platforms is challenged by incomplete, noisy social graphs and non-stationary diffusion dynamics. We propose SP-GCRL, a social-propagation-aware graph contrastive reinforcement learning framework that learns end-to-end seed selection under partial observability.We first introduce a social-propagation-aware nonlinear diffusion function to model reinforcement/diminishing effects and probability drift under repeated exposure; we then construct dual structural views and perform contrastive learning to obtain node representations robust to missing edges and weak ties, while replacing expensive strategy metrics with a GAT-based regression surrogate to improve efficiency and scalability; finally, we use DDQN to learn an end-to-end seed selection policy on top of these representations. Experiments on multiple real-world networks show that SP-GCRL achieves significant gains over heuristic and learning-based baselines across budgets and topologies, while maintaining strong large-scale scalability.
Innovation
Methods, ideas, or system contributions that make the work stand out.
graph contrastive learning
influence maximization
social propagation modeling
GAT-based surrogate
reinforcement learning
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
