Popularity Bias in Bipartite Networks: Efficiency, Errors, and Fake Views

Abstract

Popularity bias -- the tendency to make choices that are more popular -- is a widespread behavior. We incorporate this bias into a dynamic model of a bipartite economy with heterogeneous agents, where each agent primarily cares about obtaining her optimal number of partners. We provide a full characterization of steady state networks in terms of the allocation of links between the two sides of the economy. These networks, which feature a small number of hubs on the long side, refine the set of networks that form in the absence of bias, but they are not efficient in general, despite agents being rational. When irrationality (or the possibility of mistakes) in the creation and severance of links is allowed, popularity bias leads to a further refinement, as only efficient networks remain in the long run. In addition, we uncover structural conditions under which steady state networks are efficient in the absence of mistakes. We discuss empirical implications for competition and link our findings to the "Matthew effect", market share inequality, and to the emerging industry of "fake" views and reviews on social media.