REM: a learning-based framework for selecting seed users that maximize influence spread across multi-layer (multiplex) social networks. It encodes the dynamic diffusion behavior of large multiplex networks with a Propagation Mixture-of-Experts module, and treats a generative model as a reinforcement-learning policy that autonomously generates seed sets and learns to improve them. This removes prior learning-based methods' dependence on high-quality training samples and their inability to generalize to unknown diffusion patterns, improving influence spread, scalability, and inference time over the state of the art.
Proposes REM (Reinforced Expert Maximization), a learning-based framework for Multiplex Influence Maximization — selecting seed users that maximize influence spread across multi-layer social networks…