Beyond Structure: A Causal Perspective on Network Influence

Abstract

Identifying influential nodes in complex networks is a critical challenge across domains ranging from social media analytics to epidemiology and marketing. Traditional influence maximization algorithms often fall short because they rely primarily on the network structure, overlooking the intrinsic qualities of individual nodes and causal relationships that drive influence propagation. We introduce a novel Propensity Score-based Influence Maximization algorithm (PropIM) that integrates causal inference with network analysis through a dual-factor model accounting for both intrinsic node qualities and network dynamics. PropIM combines propensity score matching to estimate individual treatment effects for node intrinsic activation probabilities, combined with a weighted influence propagation model that employs a heterogeneous mechanism that reflects the variable impact of nodes during diffusion. We evaluated PropIM using academic citation networks and real-world Instagram social graphs. The experimental results show that PropIM consistently selects nodes with substantially higher real-world influence than structure-based baselines achieving up to 240% improvement in real-world influence metrics (e.g., total citations or follower count) at a smaller seed set size ( $k=10$ ), and maintains a 32.5% advantage over degree-based greedy selection at the largest seed set size ( $k=20$ ). PropIM also reduces execution time by up to 85.6% compared with unconstrained greedy search, with reductions exceeding 73% across all evaluated seed set sizes. These results demonstrate that causal node selection identifies genuinely influential nodes that structural centrality alone cannot capture.