Neural node embeddings have recently emerged as a powerful representation forsupervised learning tasks involving graph-structured data. We leverage thisrecent advance to develop a novel algorithm for unsupervised communitydiscovery in graphs. Through extensive experimental studies on simulated andreal-world data, we demonstrate that the proposed approach consistentlyimproves over the current state-of-the-art. Specifically, our approachempirically attains the information-theoretic limits for community recoveryunder the benchmark Stochastic Block Models for graph generation and exhibitsbetter stability and accuracy over both Spectral Clustering and Acyclic BeliefPropagation in the community recovery limits.
展开▼