AnomalyDAE: Dual autoencoder for anomaly detection on attributed networks

Haoyi Fan, Fengbin Zhang, Zuoyong Li

Abstract

Anomaly detection on attributed networks aims at finding nodes whose patterns deviate significantly from the majority of reference nodes, which is pervasive in many applications such as network intrusion detection and social spammer detection. However, most existing methods neglect the complex cross-modality interactions between network structure and node attribute. In this paper, we propose a deep joint representation learning framework for anomaly detection through a dual autoencoder (AnomalyDAE), which captures the complex interactions between network structure and node attribute for high-quality embeddings. Specifically, AnomalyDAE consists of a structure autoencoder and an attribute autoencoder to learn both node embedding and attribute embedding jointly in latent space. Moreover, attention mechanism is employed in structure encoder to learn the importance between a node and its neighbors for an effective capturing of structure pattern, which is important to anomaly detection. Besides, by taking both the node embedding and attribute embedding as inputs of attribute decoder, the cross-modality interactions between network structure and node attribute are learned during the reconstruction of node attribute. Finally, anomalies can be detected by measuring the reconstruction errors of nodes from both the structure and attribute perspectives. Extensive experiments on real-world datasets demonstrate the effectiveness of the proposed method.

Introduction

In the vast, interconnected webs of social media, communication logs, and e-commerce platforms, "anomalies" are more than just statistical outliers—they are often signs of network intrusions, fraudulent accounts, or system failures. Detecting these anomalies is a high-stakes task performed on attributed networks, where each node (e.g., a user) possesses a set of features (e.g., profile descriptions) alongside its structural connections.

An anomaly in such a network can take several forms: structural anomalies (unusual connectivity), attribute anomalies (wildly different feature values), or joint anomalies where attributes do not align with structural positions. Existing methods often struggle because they treat structure and attributes as separate silos or use shallow learning techniques that fail to capture complex "cross-modality" nuances.

Methodology: The AnomalyDAE Framework

The proposed AnomalyDAE is a deep joint representation learning framework that utilizes a dual autoencoder architecture to explicitly model the interactions between network structure and node attributes.

Architecture of AnomalyDAE showing the structure and attribute autoencoders.

Dual Autoencoder Architecture

The framework consists of two parallel autoencoders that interact to learn a unified representation:

  • Structure Autoencoder: This component employs a Graph Attention mechanism. Rather than treating all neighbors equally, the encoder learns the relative importance of neighbors for defining a node’s structural context. It then attempts to reconstruct the network's adjacency matrix.
  • Attribute Autoencoder: This maps raw node features into a low-dimensional latent space.
  • Joint Decoder: To reconstruct a node's attributes, the model combines the node embedding (from the structure side) and the attribute embedding. This forces the model to learn how structure and attributes interact, capturing the reality that a node's features are often a function of its community.

Anomaly Detection

The model is trained to minimize reconstruction error. If the model cannot accurately reconstruct a node's connections or its attributes based on the patterns learned from the rest of the network, that node is assigned a high anomaly score.

Experimental Evaluation

The authors evaluated AnomalyDAE on three real-world benchmarks: BlogCatalog (social network), Flickr (image-sharing), and ACM (citation network). Performance was measured using the AUC score (Area Under the ROC Curve).

Results and Discussion

AnomalyDAE consistently outperformed all baselines, including recent deep learning competitors like "Dominant."

Comparison of AUC scores across different methods and datasets.

As shown in the results, AnomalyDAE achieved significant gains over the state-of-the-art. On the BlogCatalog dataset, it reached an AUC of 97.81, compared to 78.13 for Dominant. Across the three datasets, AnomalyDAE showed improvements of 19.68 (BlogCatalog), 22.32 (Flickr), and 15.11 (ACM) percentage points over the previous best deep learning method. These results underscore the importance of modeling non-linear relationships and cross-modality interactions.

Conclusion

AnomalyDAE represents a significant step forward in anomaly detection for attributed networks. By leveraging a dual autoencoder structure and attention mechanisms, it effectively captures the complex interplay between what a node "says" and who it "talks to." While future work is needed to address scalability for massive, dynamic graphs, AnomalyDAE provides a powerful tool for identifying outliers in complex networked data.

Reference: Fan, H., Zhang, F., & Li, Z. (2020). AnomalyDAE: Dual autoencoder for anomaly detection on attributed networks. arXiv:2002.03665v2 [cs.LG].

Summary updated

Abstract

Anomaly detection on attributed networks aims at finding nodes whose patterns deviate significantly from the majority of reference nodes, which is pervasive in many applications such as network intrusion detection and social spammer detection. However, most existing methods neglect the complex cross-modality interactions between network structure and node attribute. In this paper, we propose a deep joint representation learning framework for anomaly detection through a dual autoencoder (AnomalyDAE), which captures the complex interactions between network structure and node attribute for high-quality embeddings. Specifically, AnomalyDAE consists of a structure autoencoder and an attribute autoencoder to learn both node embedding and attribute embedding jointly in latent space. Moreover, attention mechanism is employed in structure encoder to learn the importance between a node and its neighbors for an effective capturing of structure pattern, which is important to anomaly detection. Besides, by taking both the node embedding and attribute embedding as inputs of attribute decoder, the cross-modality interactions between network structure and node attribute are learned during the reconstruction of node attribute. Finally, anomalies can be detected by measuring the reconstruction errors of nodes from both the structure and attribute perspectives. Extensive experiments on real-world datasets demonstrate the effectiveness of the proposed method.
BibTeX
@inproceedings{fan2020anomalydae,
  author = {Haoyi Fan and Fengbin Zhang and Zuoyong Li},
  title = {{AnomalyDAE: Dual autoencoder for anomaly detection on attributed networks}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2020},
  url = {https://openreview.net/forum?id=cr7redkQOv},
}
Figure