Identifiability and estimation of mixed membership stochastic blockmodels

The Mixed Membership Stochastic Blockmodel (MMSB) is a widely used method for detecting overlapping communities in large network data. However, MMSB is known to be unidentifiable, which presents a challenge for practical use. Previous approaches to MMSB identifiability rely on pure nodes, or nodes belonging to only one community, which is often too restrictive for real-world applications. In this paper, we propose a new, less restrictive set of identifiability conditions for MMSB and introduce an estimator based on a specific integrated likelihood with a penalty for volume. Our proposed estimator is demonstrated to have desirable asymptotic properties and can be efficiently computed using an MCMC-EM algorithm. We illustrate the benefits of our method through simulation studies and real data applications.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Statistics

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• He, Shishuang

Contributors dc:contributor

• Liang, Feng
• Yang, Yun
• Chen, Yuguo
• Liu, Jingbo

• Mixed Membership Stochastic Blockmodels
• Identifiability
• Volume Minimization
• Sufficiently Scattered
• Volume-penalized Integrated Likelihood

Statement dc:rights

• Copyright 2023 Shishuang He

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/122228

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F122228