Chi-Square Wavelet Graph Neural Networks for Heterogeneous Graph Anomaly Detection
Authors: Xiping Li, Xiangyu Dong, Xingyi Zhang, Kun Xie, Yuanhao Feng, Bo Wang, Guilin Li, Wuxiong Zeng, Xiujun Shu, Sibo Wang
Organizations: The Chinese University of Hong Kong, Hong Kong SAR, China · Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, United Arab Emirates · Tencent, WeChat Pay, Shenzhen, China
Graph Anomaly Detection (GAD) in heterogeneous networks presents unique challenges due to node and edge heterogeneity. Existing Graph Neural Network (GNN) methods primarily focus on homogeneous GAD and thus fail to address three key issues: (C1) Capturing abnormal signal and rich semantics across diverse meta-paths; (C2) Retaining high-frequency content in HIN dimension alignment; and (C3) Learning effectively from difficult anomaly samples with class imbalance. To overcome these, we propose ChiGAD, a spectral GNN framework based on a novel Chi-Square filter, inspired by the wavelet effectiveness in diverse domains. Specifically, ChiGAD consists of: (1) Multi-Graph Chi-Square Filter, which captures anomalous information via applying dedicated Chi-Square filters to each meta-path graph; (2) Interactive Meta-Graph Convolution, which aligns features while preserving high-frequency information and incorporates heterogeneous messages by a unified Chi-Square Filter; and (3) Contribution-Informed Cross-Entropy Loss, which prioritizes difficult anomalies to address class imbalance. Extensive experiments on public and industrial datasets show that ChiGAD outperforms state-of-the-art models on multiple metrics. Additionally, its homogeneous variant, ChiGNN, excels on seven GAD datasets, validating the effectiveness of Chi-Square filters. Our code is available at https://github.com/HsipingLi/ChiGAD.
Figures & tables
Figure 1. Illustration of the ChiGAD framework, comprising three modules: (a) Multi-Graph Chi-Square Filter; (b) Interactive Meta-Graph Convolution; (c) Contribution-Informed Cross-Entropy Loss.
Metric
AUROC
AUPRC
F1-macro
Recall
LL (employed)
0.9702
0.9518
0.9195
0.8947
PCA
0.6587
0.5035
0.5975
0.4687
UMAP
0.5746
0.3852
0.5393
0.3885
IsoMAP
0.5714
0.3854
0.4417
0.3784
AE
0.9597
0.9088
0.9157
0.8847
AE_10
0.9483
0.8928
0.8798
0.8296
Table 3. Comparison of results of GAD task across diverse alignment methods on ACM.
Figure 2. Comparison of results of frequency retention task across diverse alignment methods on ACM.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Dataset
#Node
#Edge
#Feat.
Anomaly %
Split % (Train/Val/Test)
Relation
Feat. Type
ACM (het)
10,942
547,872
[193,115,250,95]
32.68/32.07/35.07
40/20/40
Co-authorship
Misc. Info.
R-I (het)
2,880,722
8,781,690
[14,171,10,13]
0.35
40/20/40
Account Interaction
Misc. Info.
R-II (het)
6,168,746
15,576,508
[14,171,10,13]
0.55
40/20/40
Account Interaction
Misc. Info.
Reddit
10,984
168,016
64
3.3
40/20/40
Under Same Post
Text Embedding
Tolokers
11,758
519,000
10
21.8
50/25/45
Work Collaboration
Misc. Info.
Amazon
11,944
4,398,392
25
9.5
70/10/20
Review Correlation
Misc. Info.
Appendix
Table 4. Summary of datasets.
Hyper-parameter
ACM
R-I
R-II
Reddit
Tolokers
Amazon
T-Finance
Questions
DGraph-Fin
T-Social
learning rate
0.0001
0.006
0.006
0.0003
0.003
0.0003
0.003
0.0001
0.01
0.01
dropout rate
0
0
0
0
0.3
0.6
0.6
0.4
0.4
0
hidden dimension
512
25
48
64
64
64
80
400
32
20
aligned dimension
512
25
48
-
-
-
-
-
-
-
epochs
200
200
200
1700
700
1550
300
150
1200
4000
MLP layers
4
2
2
2
2
3
3
4
2
3
Appendix
Table 5. Hyper-parameter setting of ChiGNN for experiments.
i
E[fi(w)]
fi−1(max)
i
E[fi(w)]
fi−1(max)
1
0.6970
0.0000
16
1.5940
1.7638
2
0.9603
0.6667
32
1.7126
1.8779
4
1.2180
1.1992
64
1.7973
1.9339
8
1.4313
1.5556
128
1.8571
1.9600
Appendix
Table 6. Expectation and most highlighted frequency of fi(w) .
Model
AUROC
AUPRC
F1-macro
Recall
ChiGAD
0.9702
0.9518
0.9195
0.8947
w/o Multi-Graph Chi-Square Filter
0.9538
0.9236
0.8889
0.8499
w/o Interactive Meta-Graph Chi-Square Convolution
0.8969
0.8410
0.7988
0.7343
w/o Contribution-Informed Cross-Entropy Loss
0.9237
0.8449
0.8526
0.8070
Variant of ChiGAD (Beta filter)
0.9411
0.9093
0.8735
0.8371
Appendix
Table 7. Experimental results of ablation study on the ACM dataset.
Graph anomaly detection (GAD) aims to identify irregular nodes or structures in attributed graphs. Neighbor information, which reflects both structural connectivity and attribute consistency with surrounding nodes, is essential for distinguishing anomalies from normal patterns. Although recent graph neural network (GNN)-based methods incorporate such information through message passing, they often fail to explicitly model its effect or interaction with attributes, limiting detection performance. This work introduces NeiGAD, a novel plug-and-play module that captures neighbor information through spectral graph analysis. Theoretical insights demonstrate that eigenvectors of the adjacency matrix encode local neighbor interactions and progressively amplify anomaly signals. Based on this, NeiGAD selects a compact set of eigenvectors to construct efficient and discriminative representations. Experiments on eight real-world datasets show that NeiGAD consistently improves detection accuracy and outperforms state-of-the-art GAD methods. These results demonstrate the importance of explicit neighbor modeling and the effectiveness of spectral analysis in anomaly detection. Code is available at: https://github.com/huafeihuang/NeiGAD.
Qing Qing, Huafei Huang, Mingliang Hou +2
College of Computer Science and Technology, Jilin University, Changchun 130012, China · School of Computer Science and Information Technology, Adelaide University, SA 5095, Australia · Guangdong Institute of Smart Education, Jinan University, Guangzhou 510632, China +1
Graph anomaly detection (GAD) aims to identify nodes that deviate from normal patterns in structure or features. While recent GNN-based approaches have advanced this task, they struggle with two major challenges: 1) homophily disparity, where nodes exhibit varying homophily at both class and node levels; and 2) limited scalability, as many methods rely on costly whole-graph operations. To address them, we propose SAGAD, a Scalable and Adaptive framework for GAD. SAGAD precomputes multi-hop embeddings and applies reparameterized Chebyshev filters to extract low- and high-frequency information, enabling efficient training and capturing both homophilic and heterophilic patterns. To mitigate node-level homophily disparity, we introduce an Anomaly Context-Aware Adaptive Fusion, which adaptively fuses low- and high-pass embeddings using fusion coefficients conditioned on Rayleigh Quotient-guided anomalous subgraph structures for each node. To alleviate class-level disparity, we design a Frequency Preference Guidance Loss, which encourages anomalies to preserve more high-frequency information than normal nodes. SAGAD supports mini-batch training, achieves linear time and space complexity, and drastically reduces memory usage on large-scale graphs. Theoretically, SAGAD ensures asymptotic linear separability between normal and abnormal nodes under mild conditions. Extensive experiments on 10 benchmarks confirm SAGAD's superior accuracy and scalability over state-of-the-art methods.
Yunhui Liu, Qizhuo Xie, Yinfeng Chen +4
State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China · National Engineering Laboratory for Big Data Analysis and Applications, Peking University, Beijing, China
In recent years, graph anomaly detection (GAD) based on frequency-domain filtering have achieved promising results. However, existing approaches still face three major challenges: First, they use static basic function to constructed graph filter which cannot effectively adapt to the frequency-domain distribution of graph data. Second, they fail to adequately consider the importance information of each attribute in the node feature vector, leading to the loss of fine-grained information. Third, they insufficiently utilize node labels for GAD. To address these issues, this paper proposes a novel graph anomaly detection method called JPGFN (Feature Transformation Enhanced Jacobi Polynomial Graph Filtering Network). First, a Feature Separation Transformation Network (FSTNN) is developed to better learn fine-grained node features by feature separation and applying nonlinear transformations to node features across different dimensions. Second, an adaptive Jacobi polynomial graph filtering module is constructed based on Jacobi polynomials to adaptively capture complex frequency-domain features of graph signals. Finally, a node label constraint module is developed to facilitate the use of node labels and enhance the performance of GAD. Experimental results on multiple real-world datasets demonstrate that the proposed method significantly outperforms mainstream approaches.
Xiang Wang, Zhijun Cheng, Zhenyu Meng
School of Artificial Intelligence and Transportation Engineering, Fujian University of Technology, Fuzhou, China · Institute of Artificial Intelligence, Fujian University of Technology, Fuzhou, China