FairNet: A Genetic Framework to Reduce Marginalization in Social Networks

Discrimination in social networks often assumes the form of marginalization against nodes with specific features, e.g., segregation of/against minorities. In this work, we propose a metric that proxies social discrimination based on salient node features in a social network. Under the assumption that in a fair social system, all individuals should be enclosed in similar social circles representing the network in its entirety, our metric assigns a marginalization score to each node in the network, identifying if they are marginalized by similar nodes (e.g., a man marginalized by other men), by different nodes (e.g., a man marginalized by women), or not marginalized at all (i.e., the node has a fair neighborhood). Moreover, we introduce FairNet, a two-fold framework that aims to reduce network marginalization in partially- and fully-attributed networks by employing genetic algorithms. We evaluate our framework on networks emerging from online social interactions and find that the two components of FairNet are able to consistently reduce marginalization.