Inequality and inequity in network-based ranking and recommendation algorithms

Abstract

Though algorithms promise many benefits including efficiency, objectivity and accuracy, they may also introduce or amplify biases. Here we study two well-known algorithms, namely PageRank and Who-to-Follow (WTF), and show to what extent their ranks produce inequality and inequity when applied to directed social networks. To this end, we propose a directed network model with preferential attachment and homophily (DPAH) and demonstrate the influence of network structure on the rank distributions of these algorithms. Our main findings suggest that (i) inequality is positively correlated with inequity, (ii) inequality is driven by the interplay between preferential attachment, homophily, node activity and edge density, and (iii) inequity is driven by the interplay between homophily and minority size. In particular, these two algorithms reduce, replicate and amplify the representation of minorities in top ranks when majorities are homophilic, neutral and heterophilic, respectively. Moreover, when this representation is reduced, minorities may improve their visibility in the rank by connecting strategically in the network. For instance, by increasing their out-degree or homophily when majorities are also homophilic. These findings shed light on the social and algorithmic mechanisms that hinder equality and equity in network-based ranking and recommendation algorithms.

Link recommendations: Their impact on network structure and minorities

“…In addition to these topological effects, social recommendations may also exacerbate the under-representation of certain demographic groups in the network. For instance, [14,15,33] show how the visibility of minorities can be amplified or mitigated by different levels of homophily within groups when using recommendation algorithms on scale-free networks. These inequalities have been also studied over time but only recently.…”

“…Synthetic networks: In order to systematically create networks as defined in Section 3.1, we employed the DPAH model [14]. This model allows to generate scale-free bi-populated directed networks with adjustable homophily (for each group), minority size, node activity, and edge density.…”

“…In-group links: As we have seen previously, the visibility of the minority can be affected by different factors, including the initial homophily of the network. Since homophily depends on the mixing of types of edges (see [14] for a detailed derivation of homophily in DPAH networks), we further investigate the evolution of in-group links over time, see Figure 7. Here, we found two main patterns.…”

“…In recent years, much attention has been paid to understanding when, and to what extent, such biases are being amplified. As an example, [15] and [14] have studied the correlation between network structure and the output of ranking algorithms in social networks. While these studies highlight that homophily-the tendency to connect to similar others-and preferential attachment-the tendency to connect to those that are already well-connected-are important structural factors that impact the visibility of nodes in algorithmic rankings, they do not compare effects over time.…”

“…Approach: We systematically compare five recommendation algorithms and apply their recommendations to several synthetic networks. We focus on scale-free directed networks with adjustable homophily and minority group size [14] and quantify the global changes in network structure, as well as the changes in connectivity for the minority group over time. In particular, we assess whether certain types of links are created more often than others and whether the network becomes more cohesive or segregated.…”