Modularity-Based Fairness in Community Detection

In this paper, we study the fairness of community structures in networks from a group-based perspective. Specifically, we assume that individuals in a social network belong to different groups based on the value of one of their sensitive attributes, such as their age, gender, or race, and we consider community fairness towards the protected group. Most previous work has focused on a balanced-based definition of fairness that seeks for an appropriate representation of the members of the protected group in each community. We introduce a novel form of community fairness, termed modularity-based fairness, that asks that the members of the protected group are well connected in their respective communities. We present results of the balanced-based and modularity-based fairness of several real and synthetic networks.