Collaborative Spectral Clustering in Attributed Networks

• Activity Number: 184 - SPEED: Variable Selection and Networks
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 11:35 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Xiaodong Jiang* and Pengsheng Ji
• Companies: and University of Georgia
• Keywords: Network Data Analysis ; Spectral Clustering ; Community Detection ; Attributed Networks ; Stochastic Block Model

Abstract

We proposed a novel collaborative spectral clustering algorithm for attributed networks, where assuming n nodes splitting into R non-overlapping communities and each node has a p-dimensional meta covariate. The connectivity matrix W is constructed with the structural adjacent matrix A and the covariate matrix X, where W = (1 ? ?)A + ?K(X, X?), ? ? [0, 1] is a tuning parameter to balance between structural and covariate information, while K is a Kernel function, i.e., a similarity function over pairs of nodes in the covariate matrix. We then perform the classical k?means clustering algorithm on the element-wise ratio matrix of the first R leading eigenvectors of W. Both theoretical and simulation studies showed the consistent performance under Stochastic Block Model (SBM), especially in imbalanced networks where most community detection algorithms fail.