Agglomerative hierarchical kernel spectral clustering for large scale networks

We propose an agglomerative hierarchical kernel spectral clustering (AH-KSC) model for large scale complex networks. The kernel spectral clustering (KSC) method uses a primal-dual framework to build a model on a subgraph of the network. We exploit the structure of the projections in the eigenspace to automatically identify a set of distance thresholds. These thresholds lead to the different levels of hierarchy in the network. We use these distance thresholds on the eigen-projections of the entire network to obtain a hierarchical clustering in an agglomerative fashion. The proposed approach locates several levels of hierarchy which have clusters with high modularity (Q) and high adjusted rand index (ARI) w.r.t. the groundtruth communities. We compare AH-KSC with 2 stateof-the-art large scale hierarchical community detection techniques.