Overlapping and nonoverlapping models
This work addresses a specific identifiability issue in network modeling for researchers in statistics or network science, but it is incremental as it builds on prior constraints for overlapping properties.
The paper tackles the problem of modeling directed networks with overlapping row communities and non-overlapping column communities, proposing a model that is identifiable when the number of row communities is less than or equal to the number of column communities, and provides spectral algorithms with consistent estimation guarantees.
Consider a directed network with $K_{r}$ row communities and $K_{c}$ column communities. Previous works found that modeling directed networks in which all nodes have overlapping property requires $K_{r}=K_{c}$ for identifiability. In this paper, we propose an overlapping and nonoverlapping model to study directed networks in which row nodes have overlapping property while column nodes do not. The proposed model is identifiable when $K_{r}\leq K_{c}$. Meanwhile, we provide one identifiable model as extension of ONM to model directed networks with variation in node degree. Two spectral algorithms with theoretical guarantee on consistent estimations are designed to fit the models. A small scale of numerical studies are used to illustrate the algorithms.