Influence Maximization on Temporal Networks with Persistent and Reactive Behaviors

Influence Maximization (IM) aims to identify a set of \(k\) seed nodes in a network that maximizes information spread, with applications in viral marketing, epidemic containment, and behavioral targeting. While most prior work assumes static networks, real-world networks are inherently temporal and shaped by user behaviors such as temporary disengagement, reactivation, and reinforcement from repeated exposures. To address these dynamics, we introduce cpSI-R , a novel diffusion model that jointly captures temporal evolution and behavioral persistence. The model allows nodes to temporarily deactivate and later reactivate under peer influence, while also reinforcing adoption likelihood through repeated interactions. We prove that the resulting influence spread function is monotone and submodular, making it amenable to efficient greedy approximation. Building on this model, we develop an efficient influence maximization framework featuring a structure-aware temporal sampling technique. We validate cpSI-R through empirical comparisons with classical diffusion models, demonstrating its consistent ability to achieve greater influence spread across temporal snapshots. Comprehensive experiments on various datasets show that our strategy significantly outperforms state-of-the-art baselines in both influence spread and runtime, with particularly strong gains on large-scale networks. These results highlight the practical utility of cpSI-R and our scalable IM framework in modeling and leveraging realistic diffusion dynamics over evolving social systems.