Fair Influence Maximization with Reverse Influence Sampling Boosted Multi-Objective Genetic Algorithm

Abstract

Influence maximization (IM) selects a small set of seed users to maximize expected diffusion in a social network, typically under the Independent Cascade model. Optimizing only global spread can amplify pre-existing structural inequities: some groups (e.g., demographics, communities, or departments) may receive far less exposure than others even when the overall spread is high. We present FIMMOGA++, a fairness-aware influence maximization framework that formulates IM as a multi-objective optimization problem over expected spread and multiple groupfairness objectives. FIMMOGA++ integrates efficient influence estimation via Reverse Influence Sampling (RIS) with a manyobjective genetic algorithm. Our framework returns a Pareto front of seed sets, explicitly exposing the trade-off between diffusion efficiency and fairness. We show that FIMMOGA++ improves fairness metrics (e.g., Max–Min group coverage and inequality) while remaining competitive in terms of spread and runtime relative to standard IM baselines.