Stealthy graph backdoor attack based on feature trigger

Abstract Recent studies have shown that Graph Neural Networks (GNNs) are vulnerable to backdoor attacks. Embedding malicious triggers (e.g., subgraphs or features) in the graph leads to erroneous outputs. Most graph backdoor attacks focus only on the effectiveness of the attack and ignore stealth, w...

Full description

Saved in:
Bibliographic Details
Main Authors: Yang Chen, Zhou Bin, Haixing Zhao
Format: Article
Language:English
Published: Springer 2025-06-01
Series:Complex & Intelligent Systems
Subjects:
Online Access:https://doi.org/10.1007/s40747-025-01934-5
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849388667150270464
author Yang Chen
Zhou Bin
Haixing Zhao
author_facet Yang Chen
Zhou Bin
Haixing Zhao
author_sort Yang Chen
collection DOAJ
description Abstract Recent studies have shown that Graph Neural Networks (GNNs) are vulnerable to backdoor attacks. Embedding malicious triggers (e.g., subgraphs or features) in the graph leads to erroneous outputs. Most graph backdoor attacks focus only on the effectiveness of the attack and ignore stealth, which can easily be detected by defense models leading to attack failure. To solve this problem, we propose a novel graph Backdoor Attack based on Feature Trigger (BAFT). Specifically, BAFT contains two modules: (1). trigger generation and embedding, (2). graph structure reconstruction and optimization. To enhance the stealthiness of the trigger, we use statistical sampling of the target label node features and select the features with the number of occurrences as the trigger. We use the node feature encoding of poisoned graphs as an approximate solution to the Singular Value Decomposition (SVD) for graph reconstruction. This approach effectively removes useless or harmful edges, thereby enhancing the homogeneity of the nodes. Then, BAFT uses optimization constraints to ensure the invisibility of the attack. The effectiveness of our proposed model is demonstrated with extensive experimental results in a node classification task. In Polblogs, Cora and Citeseer, BAFT achieves the highest attack success rate of 83.19 $$\%$$ % , 88.95 $$\%$$ % and 87.11 $$\%$$ % , respectively. At the same time, BAFT does not affect the classification accuracy of GNNs at clean nodes.
format Article
id doaj-art-929a72105d66418382dffddc0cc6a566
institution Kabale University
issn 2199-4536
2198-6053
language English
publishDate 2025-06-01
publisher Springer
record_format Article
series Complex & Intelligent Systems
spelling doaj-art-929a72105d66418382dffddc0cc6a5662025-08-20T03:42:11ZengSpringerComplex & Intelligent Systems2199-45362198-60532025-06-0111811510.1007/s40747-025-01934-5Stealthy graph backdoor attack based on feature triggerYang Chen0Zhou Bin1Haixing Zhao2School of Computer Science and Technology, Shandong Technology and Business UniversitySchool of Computer Science, Qinghai Normal UniversitySchool of Computer Science, Qinghai Normal UniversityAbstract Recent studies have shown that Graph Neural Networks (GNNs) are vulnerable to backdoor attacks. Embedding malicious triggers (e.g., subgraphs or features) in the graph leads to erroneous outputs. Most graph backdoor attacks focus only on the effectiveness of the attack and ignore stealth, which can easily be detected by defense models leading to attack failure. To solve this problem, we propose a novel graph Backdoor Attack based on Feature Trigger (BAFT). Specifically, BAFT contains two modules: (1). trigger generation and embedding, (2). graph structure reconstruction and optimization. To enhance the stealthiness of the trigger, we use statistical sampling of the target label node features and select the features with the number of occurrences as the trigger. We use the node feature encoding of poisoned graphs as an approximate solution to the Singular Value Decomposition (SVD) for graph reconstruction. This approach effectively removes useless or harmful edges, thereby enhancing the homogeneity of the nodes. Then, BAFT uses optimization constraints to ensure the invisibility of the attack. The effectiveness of our proposed model is demonstrated with extensive experimental results in a node classification task. In Polblogs, Cora and Citeseer, BAFT achieves the highest attack success rate of 83.19 $$\%$$ % , 88.95 $$\%$$ % and 87.11 $$\%$$ % , respectively. At the same time, BAFT does not affect the classification accuracy of GNNs at clean nodes.https://doi.org/10.1007/s40747-025-01934-5Graph neural networksBackdoor attackFeature triggerNode classification task
spellingShingle Yang Chen
Zhou Bin
Haixing Zhao
Stealthy graph backdoor attack based on feature trigger
Complex & Intelligent Systems
Graph neural networks
Backdoor attack
Feature trigger
Node classification task
title Stealthy graph backdoor attack based on feature trigger
title_full Stealthy graph backdoor attack based on feature trigger
title_fullStr Stealthy graph backdoor attack based on feature trigger
title_full_unstemmed Stealthy graph backdoor attack based on feature trigger
title_short Stealthy graph backdoor attack based on feature trigger
title_sort stealthy graph backdoor attack based on feature trigger
topic Graph neural networks
Backdoor attack
Feature trigger
Node classification task
url https://doi.org/10.1007/s40747-025-01934-5
work_keys_str_mv AT yangchen stealthygraphbackdoorattackbasedonfeaturetrigger
AT zhoubin stealthygraphbackdoorattackbasedonfeaturetrigger
AT haixingzhao stealthygraphbackdoorattackbasedonfeaturetrigger