Ignoring Directionality Leads to Compromised Graph Neural Network Explanations

Changsheng Sun; Xinke Li; Jin Song Dong

doi:10.1109/SPW67851.2025.00032

Ignoring Directionality Leads to Compromised Graph Neural Network Explanations

Changsheng Sun^*
, Xinke Li
, Jin Song Dong

^*Corresponding author for this work

Department of Data Science

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

Abstract

Graph Neural Networks (GNNs) are increasingly used in critical domains, where reliable explanations are vital for supporting human decision-making. However, the common practice of graph symmetrization discards directional information, leading to significant information loss and misleading explanations. Our analysis demonstrates how this practice compromises explanation fidelity. Through theoretical and empirical studies, we show that preserving directional semantics significantly improves explanation quality, ensuring more faithful insights for human decision-makers. These findings highlight the need for direction-aware GNN explainability in security-critical applications.

© 2025, Changsheng Sun. Under license to IEEE

Original language	English
Title of host publication	Proceedings - 46th IEEE Symposium on Security and Privacy Workshops (SPW 2025)
Editors	Marina Blanton, William Enck, Cristina Nita-Rotaru
Publisher	IEEE
Number of pages	7
ISBN (Electronic)	979-8-3315-6643-2
DOIs	https://doi.org/10.1109/SPW67851.2025.00032
Publication status	Published - 2025

Bibliographical note

Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s).

Research Keywords

Explainable AI
Post-hoc Explanations
Graph Learning
Trustworthy Systems

Access Output

10.1109/SPW67851.2025.00032

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@inproceedings{24e4e5f01a834844b10863254b8b61a8,

title = "Ignoring Directionality Leads to Compromised Graph Neural Network Explanations",

abstract = "Graph Neural Networks (GNNs) are increasingly used in critical domains, where reliable explanations are vital for supporting human decision-making. However, the common practice of graph symmetrization discards directional information, leading to significant information loss and misleading explanations. Our analysis demonstrates how this practice compromises explanation fidelity. Through theoretical and empirical studies, we show that preserving directional semantics significantly improves explanation quality, ensuring more faithful insights for human decision-makers. These findings highlight the need for direction-aware GNN explainability in security-critical applications. {\textcopyright} 2025, Changsheng Sun. Under license to IEEE",

keywords = "Explainable AI, Post-hoc Explanations, Graph Learning, Trustworthy Systems",

author = "Changsheng Sun and Xinke Li and Dong, \{Jin Song\}",

note = "Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s).",

year = "2025",

doi = "10.1109/SPW67851.2025.00032",

language = "English",

editor = "Marina Blanton and William Enck and Cristina Nita-Rotaru",

booktitle = "Proceedings - 46th IEEE Symposium on Security and Privacy Workshops (SPW 2025)",

publisher = "IEEE",

address = "United States",

}

Ignoring Directionality Leads to Compromised Graph Neural Network Explanations. / Sun, Changsheng; Li, Xinke; Dong, Jin Song.
Proceedings - 46th IEEE Symposium on Security and Privacy Workshops (SPW 2025). ed. / Marina Blanton; William Enck; Cristina Nita-Rotaru. IEEE, 2025.

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - Ignoring Directionality Leads to Compromised Graph Neural Network Explanations

AU - Sun, Changsheng

AU - Li, Xinke

AU - Dong, Jin Song

N1 - Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s).

PY - 2025

Y1 - 2025

N2 - Graph Neural Networks (GNNs) are increasingly used in critical domains, where reliable explanations are vital for supporting human decision-making. However, the common practice of graph symmetrization discards directional information, leading to significant information loss and misleading explanations. Our analysis demonstrates how this practice compromises explanation fidelity. Through theoretical and empirical studies, we show that preserving directional semantics significantly improves explanation quality, ensuring more faithful insights for human decision-makers. These findings highlight the need for direction-aware GNN explainability in security-critical applications. © 2025, Changsheng Sun. Under license to IEEE

AB - Graph Neural Networks (GNNs) are increasingly used in critical domains, where reliable explanations are vital for supporting human decision-making. However, the common practice of graph symmetrization discards directional information, leading to significant information loss and misleading explanations. Our analysis demonstrates how this practice compromises explanation fidelity. Through theoretical and empirical studies, we show that preserving directional semantics significantly improves explanation quality, ensuring more faithful insights for human decision-makers. These findings highlight the need for direction-aware GNN explainability in security-critical applications. © 2025, Changsheng Sun. Under license to IEEE

KW - Explainable AI

KW - Post-hoc Explanations

KW - Graph Learning

KW - Trustworthy Systems

UR - https://www.webofscience.com/wos/woscc/full-record/WOS:001541447800028

U2 - 10.1109/SPW67851.2025.00032

DO - 10.1109/SPW67851.2025.00032

M3 - RGC 32 - Refereed conference paper (with host publication)

BT - Proceedings - 46th IEEE Symposium on Security and Privacy Workshops (SPW 2025)

A2 - Blanton, Marina

A2 - Enck, William

A2 - Nita-Rotaru, Cristina

PB - IEEE

ER -

Ignoring Directionality Leads to Compromised Graph Neural Network Explanations

Abstract

Bibliographical note

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Cite this