A Pressure-Based Diffusion Model for Influence Maximization on Social Networks

A team of researchers has introduced a new AI model for simulating how influence spreads on social networks. Their Pressure Threshold (PT) model is a significant evolution of the established Linear Threshold (LT) model. The key innovation is that a user's (node's) ability to influence others is no longer static; it dynamically increases in proportion to the 'pressure' or influence they receive from their own activated neighbors. This creates a feedback loop where highly influenced users become super-spreaders, more accurately mirroring real-world viral dynamics. The work, accepted at the ICWSM 2026 conference, includes enhancements to the open-source CyNetDiff library for testing.

In applying this model to the Influence Maximization (IM) problem—the task of finding the best initial users to seed for maximum spread—the researchers made a critical discovery. Using greedy algorithms, the optimal seed sets selected under the PT model are distinct from those chosen under the older LT model. Their analysis, run on real-world network data, shows this effect is dramatically stronger in densely connected communities. This means strategies for launching marketing campaigns, public health messages, or information operations based on old models could be targeting the wrong people, especially on tight-knit platforms.