Deep reinforcement learning-based influence maximization for heterogeneous hypergraphs

In the field of social computing, understanding and optimizing information dissemination within social networks is crucial. To effectively model the complex higher-order relationships that exist in these networks, hypergraphs are employed. Current research predominantly centers on homogeneous hypergraphs, which encapsulate only a single type of relationship. However, real-world interactions often encompass more intricate and diverse higher-order relationships, which can be captured well by heterogeneous hypergraphs. This paper delves into the challenge of influence maximization within such heterogeneous hypergraphs. Specifically, we first introduce HLabel-LT, an innovative diffusion model designed for heterogeneous hypergraphs, which integrates multiple relationship types to enhance simulation accuracy. Furthermore, we propose a novel seed nodes selection framework, Heterogeneous Influence Maximization with Deep Reinforcement Learning (HIMH-DRL), which harnesses the unique adaptive trial-and-error learning capabilities of deep reinforcement learning to optimize influence spread in heterogeneous hypergraphs. To facilitate the training demands of this framework, we also present a new hypergraph generation model, HyperFF-Label, used to create artificially synthesized heterogeneous hypergraphs. Extensive experiments across various real-world datasets demonstrate that our methodology not only broadens the understanding of hypergraph dynamics but also markedly enhances the efficiency and effectiveness of influence propagation compared to traditional methods. This study underscores the potential of exploiting heterogeneity in hypergraphs to devise more effective information dissemination strategies in social networks.