Fairness-constrained multigroup influence maximization

• Published in: Knowledge and information systems Vol. 67; no. 4; pp. 3487 - 3511
• Main Authors: Zhang, Zizhen, Li, Deying, Wang, Yongcai, Chen, Wenping, Zhu, Yuqing
• Format: Journal Article
• Language: English
• Published: London Springer London 01.04.2025; Springer Nature B.V
• Subjects: Algorithms; Combinatorial analysis; Computer Science; Data Mining and Knowledge Discovery; Database Management; Egalitarianism; Information Storage and Retrieval; Information Systems and Communication Service; Information Systems Applications (incl.Internet); IT in Business; Maximization; Minority & ethnic groups; Neural networks; Nodes; Optimization; Regular Paper; Influence maximization; Social networks; Welfare economics; Group fairness
• ISSN: 0219-1377; 0219-3116
• DOI: 10.1007/s10115-024-02314-0

Influence maximization is a well-explored subject within network science, aiming to maximize the spread of influence from a given set of initial individuals to other nodes in the network. This concept finds applications in various fields such as viral marketing, information propagation, news dissemination, and vaccinations. However, traditional influence maximization objectives often overlook the equitable distribution of influenced nodes concerning sensitive attributes like race or gender. In this paper, we address the issue of fair influence maximization, aiming to achieve more equitable outcomes, particularly for minority groups. Our approach involves formulating the problem as the optimization of a welfare function that explicitly incorporates two crucial aspects of fairness: utility and equity. To tackle this challenge, we propose a novel neural network architecture consisting of two specialized subnetworks designed for handling combinatorial optimization problems on graphs. Our framework encompasses multiple notions of utility and fairness, including maximin egalitarian fairness, regularized maximin egalitarian fairness, and leximin fairness. Through extensive experimentation, we demonstrate that our framework is not only applicable in diverse scenarios but also competes favorably with existing algorithms. The results showcase the effectiveness and competitiveness of our approach in achieving fair influence maximization.