Community Detection Based on Higher-Order Network Topology and Node Covariates
Аннотация
Community detection is a fundamental task in network analysis, aiming to partition nodes into groups with similar connectivity patterns. Traditional methods often rely solely on edge information, neglecting other sources that can provide complementary insights. To fully utilize accessible information, a community detection method has been proposed, which integrates edges, triangle motifs, and node covariates, into a unified similarity matrix. The similarity incorporates two hyperparameters to adaptively balance the contributions from topology and covariates. Under the weighted stochastic block model, a spectral norm bound for the estimated similarity matrix and the consistency of community recovery have been established. Extensive simulations under various network regimes demonstrate that the proposed Higher-order Spectral Clustering method with node Covariates (HSCwithCov) outperforms the existing methods that use only subsets of the available information. Applications to the Lazega lawyers network and the Weddell Sea food web further confirm the practical utility and interpretability of the proposed method.