PRIM: Encoding Propagation Probability and Role-Aware Representation for Influence Maximization

Influence maximization in networks has demonstrated robust efficacy in the field of identifying a set of initial influential key nodes. Although recent machine learning-based methods have enhanced the performance on unknown graphs due to their stronger generalization, they ignored the diversified underlying information diffusion patterns in the real world. Moreover, there exist strong coupling characteristics between the network topology structure and the individual nodes when information spreads. To address the aforementioned issues, we propose a novel framework, i.e., Encoding Propagation Probability and Role-Aware Representation for Influence Maximization (PRIM). Specifically, we utilize a feature extractor module to distinguish the nodes with superior information transmission ability, which includes a graph diffusion convolution to learn multi-hop propagation probability matrix and a graph convolutional network to learn the role structural embedding. Then we input the features into a multi-head attention message passing module to rank the influential nodes. Extensive experiments are conducted to validate both the effectiveness of PRIM and its superiority compared to state-of-the-art methods.