A Comparative Study of Fairness Metrics for Modularity-based Community Detection

This thesis presents the evaluation of fairness metrics applied to modularity-based community detection algorithms. Community detection algorithms are widely used to uncover structural patterns in networks. However, these methods can inadvertently produce communities that reflect or reinforce demographic or structural biases present in the data. Fairness-aware community detection has emerged as a growing research area focused on promoting equitable outcomes across different social groups. We analyze how different fairness notions — those based on balance and those based on modularity — interact with community structure across multiple networks. Using synthetic and real-world networks, we assess the trade-offs between the fairness metrics under varying degrees of demographic skew and network topology. Our results highlight significant differences in the behavior of fairness metrics and reveal conditions in which one family of fairness metrics is better than the other. This work contributes empirical insights and guidance for integrating fairness into network analysis, with implications for social network mining and recommendation systems.