A reinforcement learning approach to edge suggestion for fair information access on social networks

Fangzheng Wang, Yanhao Wang, Jingjing Wang, Minghao Zhao

Abstract: Fairness in information access on social networks has been actively investigated in recent years. Most existing studies on this topic focus on the problem of Fair Influence Maximization (FIM), which aims to select a set of seed nodes such that a propagation campaign initiated by them has fair influence spread across different groups. Although FIM approaches can guarantee fair access to specific information, they cannot resolve the inherent disparities in information access between different groups arising from graph structures. To address this issue, we study the problem of augmenting the graph structure via edge suggestion towards fairer information access at the group level. Specifically, we formulate a new optimization problem called Fair Information Access via Edge Suggestion (FIAES) that identifies a set of at most b non-existing edges to be added into the graph such that they not only maximally increase the total influence spread, but also ensure fairness in the sense that the influence spreads within different groups are proportional to their population sizes. Since FIAES is NP-hard and cannot be approximated within any constant factor, unless P=NP, we propose FIAES-RL, a reinforcement learning-based algorithm for edge selection that strikes a balance between influence and fairness objectives. Finally, with extensive experimentation on four synthetic and real-world networks, we demonstrate that FIAES-RL outperforms several state-of-the-art baseline methods for fairness-aware edge suggestion, reducing inequity in information access while significantly boosting information propagation.