Quantifying and Mitigating Consensus Disparity in Social and Information Networks

Abstract

We introduce a computational framework to measure and optimize disparity, which corresponds to the difference in consensus outcomes attributable to distinct social groups, under classical models of opinion dynamics. We study this problem in the Friedkin-Johnsen setting under uncertainty about group structure and characterize its algorithmic complexity. For the structural analysis, we demonstrate that disparity can be arbitrarily larger than polarization in well-connected networks that nonetheless carry an identifiable group structure. For the mitigation problem, we derive robust formulations and active set optimization procedures to minimize worst-case disparity via recommendation reweighing and opinion seeding. Our methods provide provable guarantees and are validated on multiple real-world social networks. The results bridge opinion dynamics and network optimization, offering computational tools for analyzing and reducing polarization in social networks.