Anomaly Detection on Attributed Networks via Contrastive Self-Supervised Learning (2103.00113v2)

Abstract: Anomaly detection on attributed networks attracts considerable research interests due to wide applications of attributed networks in modeling a wide range of complex systems. Recently, the deep learning-based anomaly detection methods have shown promising results over shallow approaches, especially on networks with high-dimensional attributes and complex structures. However, existing approaches, which employ graph autoencoder as their backbone, do not fully exploit the rich information of the network, resulting in suboptimal performance. Furthermore, these methods do not directly target anomaly detection in their learning objective and fail to scale to large networks due to the full graph training mechanism. To overcome these limitations, in this paper, we present a novel contrastive self-supervised learning framework for anomaly detection on attributed networks. Our framework fully exploits the local information from network data by sampling a novel type of contrastive instance pair, which can capture the relationship between each node and its neighboring substructure in an unsupervised way. Meanwhile, a well-designed graph neural network-based contrastive learning model is proposed to learn informative embedding from high-dimensional attributes and local structure and measure the agreement of each instance pairs with its outputted scores. The multi-round predicted scores by the contrastive learning model are further used to evaluate the abnormality of each node with statistical estimation. In this way, the learning model is trained by a specific anomaly detection-aware target. Furthermore, since the input of the graph neural network module is batches of instance pairs instead of the full network, our framework can adapt to large networks flexibly. Experimental results show that our proposed framework outperforms the state-of-the-art baseline methods on all seven benchmark datasets.

• The paper introduces CoLA, a contrastive self-supervised framework that leverages local substructure sampling to detect anomalies in attributed networks.
• It utilizes a GNN-based model with innovative contrastive mechanisms to overcome limitations of autoencoder-based approaches, enhancing scalability.
• Evaluation on benchmark datasets confirms CoLA’s superior performance in identifying abnormal nodes across diverse network scenarios.

Anomaly Detection on Attributed Networks via Contrastive Self-Supervised Learning

The paper "Anomaly Detection on Attributed Networks via Contrastive Self-Supervised Learning" proposes an advanced framework for identifying anomalies within attributed networks by utilizing contrastive self-supervised learning techniques. The research addresses notable limitations encountered by existing methods, which typically employ graph autoencoders as their backbone and struggle to scale effectively on large networks.

The central contribution of this paper is the introduction of the CoLA framework, representing Contrastive self-supervised Learning for Anomaly detection. CoLA is designed to harness local information from network data through an innovative type of contrastive instance pair sampling. This framework efficiently captures relationships between each node and its neighboring substructure. Through an unsupervised approach, the CoLA framework extracts relevant knowledge by employing a sophisticated graph neural network (GNN)-based model and carefully crafted contrastive learning mechanisms.

A salient feature of CoLA is its ability to measure the agreement between instance pairs with distinct scores, which are then employed to assess the abnormality of each node. The contrastive learning paradigm allows the model to focus specifically on anomaly detection objectives, rather than unsupervised reconstruction tasks that autoencoder-based approaches are inclined toward. By generating batches of instance pairs, CoLA can efficiently process large networks, mitigating scalability issues.

Evaluation on multiple benchmark datasets demonstrates CoLA's effectiveness in outperforming state-of-the-art baseline methods across all scenarios tested. The numerical results underscore its enhanced performance, achieved by maximizing network data exploitation and focusing on both attribute and structural information. Such results illustrate the framework’s superiority in identifying nodes that significantly deviate from the norm, a crucial capability for applications in fraud detection, network security, and monitoring social media activities.

From a theoretical standpoint, CoLA highlights the advantages of embedding self-supervised contrastive learning within the graph learning domain, showcasing its potential for future expansion and adaptation. Practically, the framework sets a precedent for deploying scalable and efficient anomaly detection systems in real-world network environments, where high-dimensional attributes and intricate structural dynamics present substantial challenges.

Future research directions proposed by the authors may involve further enhancing the framework’s scalability, expanding its application to diverse forms of networks, and integrating it with hybrid or transfer learning paradigms to improve anomaly detection accuracy even further. Moreover, the exploration of additional contrastive learning methods and their adaptability to varying network conditions remains a compelling prospect.

Overall, this paper provides a substantial contribution to the field of anomaly detection by leveraging cutting-edge techniques in contrastive self-supervised learning, offering both theoretical insights and practical advancements in employing GNNs for complex network analyses.