Influence Strength Estimation in Hyperbolic Space for Social Influence Maximization

A research team led by Hongliang Qiao has published a new paper, "Influence Strength Estimation in Hyperbolic Space for Social Influence Maximization," introducing a novel AI model called HIM. The work tackles the classic Influence Maximization (IM) problem—finding the smallest set of users to trigger the widest information cascade in a network—but with a key innovation. Instead of using standard Euclidean geometry, HIM leverages hyperbolic space, a curved geometric representation that naturally captures the hierarchical and tree-like structure of social influence. This allows the model to encode a user's potential influence spread directly into their position in this space, with more influential users clustering closer to the origin.

Traditional IM methods rely on knowing the exact parameters of a diffusion model (like how likely a user is to share a post), which is often unknown in real-world scenarios. HIM is model-agnostic; it learns influence patterns directly from network structure and historical propagation data. The system has two core components: a hyperbolic influence representation module that creates the user embeddings, and an adaptive seed selection module that picks the optimal influencers. In extensive experiments across five network datasets, HIM demonstrated superior effectiveness and efficiency, highlighting its potential for scalable application in real-world social media marketing and information campaign planning where the rules of influence are complex and latent.