Inequality and inequity in network-based ranking and recommendation algorithms

Lisette Espín-Noboa, Claudia Wagner, Markus Strohmaier, Fariba Karimi

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. These findings shed light on the social and algorithmic mechanisms that hinder equality and equity in network-based ranking and recommendation algorithms.

Introduction

Ranking algorithms like PageRank and Twitter’s "Who-to-Follow" (WTF) are often perceived as objective arbiters of importance. However, these algorithms operate on social networks that are inherently biased. This research explores the gap between inequality (the "rich-get-richer" distribution of scores among individuals) and inequity (the systematic under-representation of minority groups in top results). The central question is how the underlying structure of connections—specifically homophily—dictates whether an algorithm will be fair or biased.

Methodology: The DPAH Model

To study these dynamics, the authors developed the Directed Preferential Attachment with Homophily (DPAH) model. This model incorporates three key mechanisms:

  1. Preferential Attachment: Popular nodes are more likely to receive new links.
  2. Homophily: Nodes tend to connect with others sharing similar attributes (e.g., gender or political leaning).
  3. Directionality: The model accounts for directed links (e.g., "following"), which is essential for social media and citation networks.

By adjusting these parameters, the researchers simulated various network "societies" to observe how PageRank and WTF distribute visibility.

Key Insights: A Taxonomy of Bias

The study reveals that the visibility of a minority group is primarily dictated by the behavior of the majority. The researchers categorized the outcomes into a taxonomy based on the Gini coefficient of rank distribution (inequality) and the level of minority representation (inequity).

A taxonomy of algorithmic outcomes based on inequality and inequity.

Key findings include:

  • Inequality and Inequity are Linked: Highly skewed rankings where a few individuals dominate tend to be the most unfair to minority groups.
  • The "Glass Ceiling" is Structural: Inequity is remarkably consistent across top ranks. If a minority is under-represented in the top 5%, they are typically under-represented in the top 20% and 50% as well. Simply "expanding the search" doesn't fix the bias.
  • Strategic Connecting: Minorities can potentially overcome bias by increasing their out-degree or homophily, but this is highly dependent on the majority's behavior.

Results and Empirical Validation

The researchers tested their model against four real-world datasets: citation networks (APS), Twitter retweet networks (Hate), political blog hyperlinks (Blogs), and Wikipedia politician pages.

Relationship between network homophily, ranking inequality, and minority inequity across different scenarios.

As shown in the figure above, the degree of homophily significantly impacts outcomes:

  • Homophilic Majorities: When the majority prefers their own kind, minority representation is systematically reduced in the rankings.
  • Neutral Majorities: Minority representation is replicated, matching their actual population size.
  • Heterophilic Majorities: Minorities can actually be amplified to the top of the rankings.

The DPAH model accurately explained disparities in the Blogs and Wikipedia networks, where both popularity and group-preference are active. In contrast, citation networks were primarily driven by preferential attachment alone.

Conclusion

This work highlights that algorithmic "fairness" cannot be achieved by looking at the algorithm in isolation; the social structure it processes is a primary driver of bias. While the current model focuses on binary attributes and uniform behavior, it provides a foundational framework for understanding how network-based ranking algorithms can hinder equality and equity. Future work should address intersectionality and develop automated de-biasing tools based on these structural insights.

Summary updated

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.
BibTeX
@inproceedings{espinnoboa2022inequality,
  author = {Lisette Espín-Noboa and Claudia Wagner and Markus Strohmaier and Fariba Karimi},
  title = {{Inequality and inequity in network-based ranking and recommendation algorithms}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-05434-1},
}
Figure