Dynamic graph neural network framework for real-time multi-modal data analysis and predictive modeling

In recent years, Graph Neural Networks (GNNs) have become increasingly prominent for analyzing complex, interconnected data across fields such as transportation, social networks, and cybersecurity. Despite their advancements, many existing GNN models struggle to capture the intricate interactions among temporal, spatial, and domain-specific knowledge, particularly as these factors evolve dynamically, while also accounting for the complexities of multi-modal data in real time, with current GNN architectures often falling short in leveraging cross-modal correlations. We present a novel Dynamic Graph Neural Networks (DGNNs) Framework that integrates Partial Differential Equations (PDEs), temporal-spatial modeling, and domain-specific knowledge to address these gaps. By enabling real-time processing of multi-modal data, this framework bridges real-world dynamic systems with the evolving landscape of AI and machine learning applications. This interdisciplinary approach uniquely advances AI, machine learning, and big data analytics by harmonizing spatial-temporal dynamics, domain customization, and multi-modality integration in a cohesive framework. GNNs have become essential tools for analyzing complex, interconnected data in domains such as transportation, social networks, and cybersecurity. However, current GNN models often struggle to effectively capture the dynamic interactions of temporal, spatial, and domain-specific knowledge, especially when processing multi-modal data in real time. This dissertation presents the development of a DGNNs Framework designed to overcome these challenges, illustrated through extensive use cases. For instance, in traffic prediction, experiments using datasets such as Performance Measurement System Bay Area (PEMS-BAY), Metropolitan Traffic Los Angeles (METR-LA), and other PeMS Performance Measurement System datasets demonstrate the framework’s superior performance in prediction accuracy and robustness, effectively managing real-world data variability and spatial-temporal dependencies. Additionally, the framework efficiently models inter-variable dependencies in Multivariate Time Series (MTS) forecasting in domains such as energy, weather, and environmental monitoring, achieving stable long-horizon predictions through its PDE-enhanced graph structure. Ultimately, the framework’s capabilities extend to social media analysis for misinformation detection and rumor spread pattern discovery, with superior classification results on datasets like Pheme, Twitter15, Twitter16, and WEIBO. These examples showcase how the framework uncovers evolving patterns across platforms by processing multi-modal data inputs such as text and network interactions, surpassing traditional models.

Name thesis:degree_name

Ph.D. (Doctor of Philosophy)

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Computer Science (UMKC)

Grantor

University of Missouri--Kansas City

Year dc:date.issued

2025

Author dc:creator

• Almousa, Ghadah Fahad M.

Advisor dc:contributor.advisor

• Lee, Yugyung, 1960-

Language dc:language.iso

en_US

Handle dc:identifier.uri

https://hdl.handle.net/10355/110328

OAI identifier oai:identifier

oai:mospace.umsystem.edu:10355/110328