Open Collections

Eslami, Roozbeh

University of British Columbia

Networks have been used to model complex real-world systems where pairwise interactions between objects are captured by the edges of the network. Recently, there has been a growing interest in meso-scale network structures. These structures go beyond pair-wise interactions and include higher-order sets of nodes such as, cliques, simplexes, hyperedges, etc. Interesting phenomena have been observed when higher-order interaction is taken into account. In this thesis, we studied the impact of meso-scale structures on the behavior of social contagion systems and machine learning models for network data. Using intersection models of graphs, we proposed a superimposition model of networks to better understand the behavior of social contagion process where meso-scale interactions may play a significant role. We investigated different higher-order propagation mechanisms and the nature of phase transitions in the outcome of the spreading processes. Moreover, we did extensive simulation to validate our assumptions. We also developed a Python package for modeling and simulating spreading mechanisms on standard and higher-order graphs. Furthermore, we investigated the effects of meso-scale structures on graph neural networks and our main goal was to investigate whether it is beneficial to incorporate information from the network community structure into the deep learning model. We investigated different ways of incorporating community structure information in the deep learning model and provided a comparison to other commonly used models on a variety of datasets.

Text

2022-10-19

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/82905

Computer Science

University of British Columbia

UBCO

http://creativecommons.org/licenses/by-nc-nd/4.0/