Date of Award
2024
Document Type
Thesis (Ph.D.)
Department or Program
Computer Science
First Advisor
Soroush Vosoughi
Abstract
Graphs are ubiquitous structures that model relationships across domains, from social networks to biological systems. This thesis, titled ``Computational Insights into Graph Data: Representation, Embedding, and Applications,'' investigates advanced methodologies for analyzing graph data through innovative representation learning and embedding techniques.
A key focus is developing hyperbolic node embeddings to efficiently capture hierarchical and structural patterns in complex networks. This work introduces novel gradient computation methods in hyperbolic spaces, extends these embeddings to heterogeneous and dynamic networks, and addresses challenges in maintaining structural role proximity.
In addition, this thesis explores methods for structural role embeddings, enabling the representation of nodes with similar roles independent of their graph positions. These methods are then extended to multiplex and attributed networks, with applications to uncovering influential roles in social media.
At the graph level, we propose embedding techniques that encode global properties through diffusion-wavelet-based characterizations of node feature distributions. This approach enhances graph classification tasks, while dynamic graph-level embeddings are developed to model temporal changes effectively.
The practical utility of these methodologies is demonstrated through applications such as analyzing influential actors on social media and modeling financial market dynamics. This thesis advances graph representation learning by merging theoretical innovations with real-world applications, offering robust tools for intuitive and effective relational data analysis.
Recommended Citation
Wang, Lili, "Computational Insights into Graph Data: Representation, Embedding, and Applications" (2024). Dartmouth College Ph.D Dissertations. 330.
https://digitalcommons.dartmouth.edu/dissertations/330
