FAIRNESS-AWARE INFLUENCE MAXIMIZATION IN SOCIAL NETWORKS USING ADVERSARIALGRAPH EMBEDDINGS

Abstract

Influence Maximization (IM) is a well-known problem of Social Network Analysis dealing with finding the most influential seed nodes in order to maximize the information spread based on a particular diffusion model. Despite the fact that conventional IM methods consider only maximizing influence, modern studies demonstrate that these methods may significantly increase existing biases inherent in social networks leading to uneven coverage, discrimination against vulnerable populations, and biased results in marketing campaigns. This review offers a detailed exploration of the topic of Fairness-Aware Influence Maximization (FAIM) that includes the latest theoretical background, algorithms, and types of fairness measures proposed in recent studies. We analyze various cases of bias that may arise due to demographic differences, clustering, homophily, and exposure in equalities using ideas offered in seminal papers including Stoica and Chaintreau’s novel definition of fairness for IM [1], Fairscale approach of Feng et al. for fairness through embeddings at scale [2], fairness conditions for diffusion blocking and adversarial fair embeddings, as well as evolutionary and fairness-aware infection analyses from theses. Our findings are based on a synthesis of eight leading papers, where we propose a unified understanding of fairness constraints, trade-offs in algorithmic development, and methods of evaluation. We provide a formalization of diffusion fairness, equity-sensitive seeding, community-sensitiveness of exposure balancing, and representation preserving in massive networks. A general FAIM framework is provided, along with the analysis of existing algorithms, fairness metrics, and diffusion outcomes. Open research questions include scalability of fairness constraints, multiple at tributes, robustness to adversarial attacks, and resolution of trade-offs between fairness and utility.