Statistical Models and Algorithms for Large Network Analysis
Open Access
Author:
Vu, Duy Quang
Graduate Program:
Statistics
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
May 02, 2012
Committee Members:
David Russell Hunter, Dissertation Advisor/Co-Advisor C Lee Giles, Committee Member Debashis Ghosh, Committee Member Aleksandra B Slavkovic, Committee Member
Keywords:
network work analysis survival modeling dynamic network formation latent class modeling
Abstract:
The emergence and growth of online social media and network services has provided
us with new data collection tools to study human behaviors. Traces of these behaviors
have accumulated into a large number of network data sets, of which many are recorded
on the continuous-time scale. The main challenges in the modeling and understanding
of these online data sets are their scale, the complexity of dynamic network processes,
and the heterogeneity among actors. To address these challenges, two different statistical
modeling approaches are explored in this dissertation. For cross-sectional data,
latent class network models based on dyad independence are considered and extended.
For continuous-time data, event-based network models based on survival analysis are
further developed. Their potential applications are demonstrated through a diverse list
of modeling problems for network formation and behavior processes. Fast inference
algorithms for these models based on the sparsity property are also discussed.
