Dissertation
Model-based edge clustering for networks: theory and applications
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2024
DOI: 10.25820/etd.007558
Abstract
This dissertation explores advanced methodologies for clustering in network analysis, with a particular focus on edge clustering algorithms. Traditional network analysis has primarily centered around node-based methods, but edge clustering offers a more nuanced understanding of relationships between entities. The work introduces several novel extensions to existing edge clustering frameworks, addressing key limitations such as the neglect of edge weights and the challenge of noisy edges—random connections that obscure the underlying network structure.
The Weighted Latent Space Edge Clustering (WLSEC) model is presented as an extension of the Latent Space Edge Clustering (LSEC) model, incorporating edge weights to capture the varying strength of relationships between nodes. This model allows for a more detailed and accurate representation of networks where relationships differ in intensity, such as patient transfer networks in healthcare. The dissertation further advances this work by introducing the Weighted Edge Clustering Adjusting for Noise (WECAN) model, which identifies and mitigates the impact of noisy edges. By distinguishing between meaningful and spurious edges, WECAN enhances the accuracy and interpretability of the clustering process.
Additionally, this research applies edge clustering methodologies to real-world problems, including the multiple source detection (MSD) problem in epidemiology. The MSD problem involves identifying multiple origins of diffusion processes, such as the spread of diseases or information. The integration of edge clustering with traditional MSD frameworks leads to significant improvements in detecting multiple sources within networks.
The effectiveness of the proposed models is validated through extensive simulation studies and real-world applications, demonstrating their superiority over traditional node-based methods. This work not only provides theoretical advancements in edge clustering algorithms but also offers practical tools for analyzing complex networks in fields such as healthcare and epidemiology.
Details
- Title: Subtitle
- Model-based edge clustering for networks: theory and applications
- Creators
- Haomin Li
- Contributors
- Daniel K. Sewell (Advisor)Grant D. Brown (Committee Member)Brian J. Smith (Committee Member)Jacob J. Oleson (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Biostatistics
- Date degree season
- Autumn 2024
- DOI
- 10.25820/etd.007558
- Publisher
- University of Iowa
- Number of pages
- ix, 84 pages
- Copyright
- Copyright 2024 Haomin Li
- Language
- English
- Date submitted
- 12/08/2024
- Description illustrations
- Illustrations, tables, graphs, charts
- Description bibliographic
- Includes bibliographical references (pages 81-84).
- Public Abstract (ETD)
This dissertation introduces innovative methods to analyze complex networks, focusing on understanding relationships between different entities, such as patients in healthcare systems or individuals in social networks. Traditional approaches often concentrate on the nodes (entities) of a network, but this research emphasizes the importance of examining the connections (edges) between these nodes, providing deeper insights that conventional methods may overlook.
The work presents two key contributions. The first is the Weighted Latent Space Edge Clustering (WLSEC) model, which extends existing edge clustering methods by considering the strength of relationships. By incorporating edge weights into the analysis, WLSEC allows for more precise clustering of networks, helping to uncover subtle patterns in data, such as patient transfers between hospitals.
The second contribution is the Weighted Edge Clustering Adjusting for Noise (WECAN) model, which builds on WLSEC by also accounting for noisy or irrelevant connections that can distort the true structure of a network. WECAN provides a data-driven approach to separate meaningful connections from noise, further improving the accuracy of network analysis.
These models are applied to real-world problems, including the detection of multiple sources of disease outbreaks in epidemiology. The research demonstrates how these new methods outperform existing techniques, offering more accurate tools for analyzing complex networks in various fields.
- Academic Unit
- Biostatistics
- Record Identifier
- 9984774456302771
Metrics
4 File views/ downloads
9 Record Views