The Effect of Link Recommendation Algorithms on Network Centrality Disparities

Link recommendation algorithms are nowadays ubiquitous in online social networks. It is fundamental to understand their impact in users’ capacity to spread information and the network centrality of different groups. This paper investigates the effect of the Stochastic Approach for Link-Structure Analysis (SALSA), a popular link recommendation algorithm, on network centrality and disparate exposure in networks composed of a majority and minority group. While previous works mainly focus on exposure inequalities along degree centrality (i.e., how frequently individuals from different groups are recommended), it remains unknown how link recommendation impacts groups’ visibility along different network centrality measures. We use two different centrality metrics that capture more granular information: betweenness (BC) and closeness (CC) centrality. Resorting to synthetic networks and as in previous works, we observe that SALSA can reduce the relative degree centrality of a minority group in a network, when both groups are homophilic. A similar conclusion is observed for BC. Conversely, we observe that CC of both minority and majority groups increases after SALSA recommendations are followed. Lastly, we propose the concept of kn-Interventions to mitigate disparate exposure effects and demonstrate that this simple intervention has a positive effect on the minority’s visibility. Our findings elucidate the need to consider different measures capturing minorities’ visibility when evaluating the effects of link recommendation in social contexts.