Abstract:
|
Spectral clustering has been extensively applied in statistic network community detection under the framework of Stochastic Blockmodel (SBM). Recently, a new approach called Spectral Clustering On Ratio-of-Eigenvectors (SCORE) has been proposed, aiming to classify communities under Degree Corrected Blockmodel (DCBM). DCBM is more general than SBM because vertices degree heterogeneity is engaged. In this paper, we consider the problem of dynamic network community discovery over time under DCBM assumption. Borrowing the idea of SCORE, we are able to minimize the objective function to discover the best community structure at time t, at the same time, guarantee the smoothness of the variation of communities. Both simulated and real data are applied to validate proposed method.
|