Modeling Information Diffusion in Networks with Unobserved Links

Modeling information diffusion in networks enables reasoning about the spread of ideas, news, opinion, and technology across a network of agents. Existing models generally assume a given network structure, in practice derived from observations of agent communication or other interactions. In many realistic settings, however, observing all connections is not feasible. We consider the problem of modeling information diffusion when the network is only partially observed, and investigate two approaches. The first learns graphical model potentials for a given network structure, compensating for missing edges through induced correlations among node states. The second learns the missing connections directly. Using data generated from a cascade model with different network structures, we empirically demonstrate that both methods improve over assuming the given network is fully observed, as well as a previously proposed structure-learning technique. We further find that potential learning outperforms structure learning when given sufficient data.