Author ORCID Identifier
0009-0007-8031-661X
Document Type
Dissertation
Date of Award
5-31-2025
Degree Name
Doctor of Philosophy in Information Systems - (Ph.D.)
Department
Informatics
First Advisor
Michael J. Lee
Second Advisor
Hua Wei
Third Advisor
Yi-Fang Brook Wu
Fourth Advisor
Hai Nhat Phan
Fifth Advisor
Mark Cartwright
Sixth Advisor
Dongsheng Luo
Abstract
In the evolving landscape of artificial intelligence (AI), Graph Neural Networks (GNNs) have garnered growing prominence for their adeptness in processing graph-structured data. Despite this, the interpretability of their predictions often remains elusive. The demand for transparency and explainability in complex prediction models has reached unprecedented levels. To address this, post-hoc instance-level explanation techniques have emerged, aiming to unveil the rationale behind GNN predictions. These techniques endeavor to unearth substructures that elucidate the predictive behavior of trained GNNs.
This dissertation embarks on an exploration of Explainable AI (XAI) technologies within the realm of GNNs. Amid the challenges posed by the distribution shifting problem and the gap from classification tasks to regression tasks, as well as studying the confidence of the explanations, this research endeavors to make GNNs to be more transparent and interpretable tools.
The motivation stems from a dual mandate: the necessity for interpretability in AI systems and the potential of GNNs to decipher complex relationships inherent in graph data and be employed in many real-life applications. Delving into the intricacies of GNN decision-making, the research aims to not only address challenges but also spearhead advancements in Explainable AI for Graph Neural Networks (XAIG). This journey encompasses strategies to mitigate the impact of the distribution shifting problem in existing works, adapt GNNs from classification tasks to regression tasks, and quantify the confidence or uncertainty of explanations. The goal is to establish the XAIG technology as a mature, dependable, and extensively employed tool in current GNN applications, thus enhancing the overall utilization and effectiveness of GNNs.
By establishing a foundation for transparent and explainable GNNs, this dissertation bridges the gap between advanced AI methodologies and human comprehension. The culmination of this research promises enhanced reliability in GNN predictions while contributing to the broader discourse on AI ethics and accountability. In a future where AI decisions are credible, accessible, and aligned with societal values, XAIG emerges as a cornerstone, harmonizing AI's computational prowess with human reasoning.
Recommended Citation
Zhang, Jiaxing, "Towards explainable AI on graph neural networks: XAIG" (2025). Dissertations. 1842.
https://digitalcommons.njit.edu/dissertations/1842
