Multi-view graph clustering

• Main Author: Yap, Nicholas Guo Dong
• Other Authors: Ke Yiping, Kelly
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/175282
• Institution: Nanyang Technological University

This study explores the adaptation and application of Deep Modularity Networks (DMoN) for multi-view graph clustering, a technique crucial for extracting insights from complex networks across various domains. By integrating multiple perspectives or views of graph data, our approach, termed Multi-view Deep Modularity Networks (MVDMoN), seeks to enhance clustering performance beyond what is achievable with single-view analyses. We focus on the UCI Multiple Features dataset, leveraging its six distinct views of handwritten digits to test our model’s efficacy. Our results indicate that MVDMoN can effectively adjust the weightage of various views to optimize clustering outcomes, revealing its adaptability and potential for uncovering valuable patterns not evident when views are analyzed in isolation. The study also enhances clustering performance over its single-view counterpart, particularly in terms of Normalized Mutual Information (NMI) and F1 score. Through potential enhancements such as ablation studies and exploration of additional datasets, subsequent research can build upon our findings, advancing the field towards more nuanced and effective clustering solutions.