Optimal injection points for information diffusion

The success of economic policy often hinges on the diffusion of information on its existence and nature through social networks. Informing all potential beneficiaries directly is costly, hence policymakers may prefer to inform a few key agents instead. Existing approaches typically select such injection points based on network centrality measures, and propose informing the most centrally located agents. This does not account for overlap in information diffusion, and can inhibit diffusion to the periphery. This note proposes a method to optimally select injection points accounting for overlap in information diffusion. The approach is based on that of Banerjee et al. (2013) and involves solving a mixed integer program. For large networks, a convex relaxation of the mixed integer program is more computationally tractable.