Modeling Node Exposure for Community Detection in Networks

In community detection, datasets often suffer a sampling bias for which nodes would normally have high affinity appear to zero affinity. This happens example when two affine users of social network were not exposed one another. Community detection on this kind data suffers then from considering as affine. To solve problem, we explicitly model the (non-) exposure mechanism in Bayesian framework, by introducing set additional hidden variables. Compared approaches do exposure, our method is able better reconstruct input graph, while maintaining similar performance recovering communities. Importantly, it allows estimate probability that been exposed, possibility available with standard models.