GraphC: Hierarchical clustering of signed graph networks

Abstract

• graphC is a k-independent hierarchical clustering method for signed graphs. • It scales to large sparse signed networks with tens of millions of edges. • It avoids spectral decomposition and eigenvalue pollution issues without requiring training or hyperparameter tuning. • It treats positive and negative edges symmetrically. • It outperforms the second-best algorithm across 14 datasets by 18.6% on average.