A model for optimizing article recommendation for reducing polarization

Abstract

Online social networks have been charged with enhancing and augmenting polarization in society. This polarization can have negative repercussions on the health of both individuals and society on the whole. Hence, it is vital that our online social networks are powered by algorithms that avoid polarisation and seek to curtail it. One target would be the curated news feed supplied to users in online social networks. In this paper, we present a stochastic dynamic programming (SDP) model that seeks to recommend articles to users with the aim of simultaneously retaining their subscription whilst reducing polarization. We also present heuristics to approximate the solution alongside an empirical comparison of the optimal vis-a-vis these heuristics. We find that, in general, recommending articles that tend towards neutrality has a positive effect in reducing polarization under our SDP model.