A0497
Title: Network goodness-of-fit for the block-model family
Authors: Jingming Wang - University of Virginia (United States) [presenting]
Jiashun Jin - Carnegie Mellon University (United States)
Tracy Ke - Harvard University (United States)
Jiajun Tang - Univeristy of Pennsylvania (United States)
Abstract: The block model family is widely used in network modeling and includes four popular models: SBM, DCBM, MMSBM, and DCMM. However, the question of which block model best fits real networks has received limited attention in the literature. A novel approach is introduced using cycle count statistics to address the goodness-of-fit for these block models. By leveraging the cycle count statistics and a network fitting scheme, four GoF metrics are constructed with parameter-free limiting distributions of N(0,1) under the assumed models. These GoF-metrics are applied to some frequently-used real networks for comparison. The numerical results suggest that DCMM is particularly promising for modeling undirected networks.
