Network community detection with edge classifiers trained on LFR graphs

Graphs generated using the Lancichinetti-Fortunato-Radicchi (LFR) model are widely used for assessing the performance of network community detection algorithms. This paper investigates an laternative use of LFR graphs: as training data for learning classifiers that discriminate between edges that are ‘within’ a community and ‘between’ network communities. The LFR generator has a parameter that controls the extent to which communities are mixed, and hence harder to detect. We show experimentally that a linear edge-wise weighted support vector machine classifier trained on a graph with more mixed communities also works well when tested on easier graph instances, while it achieves mixed performance on real-life networks, with a tendency towards finding many communities.