Link recommendation algorithms are applied in online social networks to offer friend or followee recommendations. While these algorithms help users expand their networks, they can also unintentionally reinforce inequalities by affecting the visibility of certain groups.

Prior work considers fairness in link-recommenders based on individuals’ degree, i.e., the number of connections they receive as a result of being recommended. This fairness notion does not cover how recommendations shape the ability of groups to mediate and broker information flows. In this paper, we aim at auditing link-recommendation algorithms through the lens of betweenness centrality, a measure of how groups control information pathways.

We use synthetic networks with tunable levels of within- and across- group connections (i.e., asymmetric homophily), and we show that this structural property fundamentally determines whether fairness interventions reduce betweenness centrality disparities.

To directly address these inequities, we introduce Flowtune, a new method to mitigate betweenness centrality disparities in online social networks. Flowtune adapts recommendations based on nodes’ structural roles. Across synthetic and empirical networks, we show that Flowtune reduces betweenness disparity while preserving utility.

Our contribution stresses the need to audit social network algorithms according to broader fairness definitions and highlights the need to design algorithms that explicitly account for such fairness principles.

M.M. Pawar, F. Karimi, F.P. Dos Santos, Fairness Beyond Popularity: Auditing and Mitigating Betweenness Centrality Disparities in Link Recommendation, Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency (2026) 5068-5094.