A Comprehensive Survey on Graph Anomaly Detection with Deep Learning (2106.07178v5)

Abstract: Anomalies represent rare observations (e.g., data records or events) that deviate significantly from others. Over several decades, research on anomaly mining has received increasing interests due to the implications of these occurrences in a wide range of disciplines. Anomaly detection, which aims to identify rare observations, is among the most vital tasks in the world, and has shown its power in preventing detrimental events, such as financial fraud, network intrusion, and social spam. The detection task is typically solved by identifying outlying data points in the feature space and inherently overlooks the relational information in real-world data. Graphs have been prevalently used to represent the structural information, which raises the graph anomaly detection problem - identifying anomalous graph objects (i.e., nodes, edges and sub-graphs) in a single graph, or anomalous graphs in a database/set of graphs. However, conventional anomaly detection techniques cannot tackle this problem well because of the complexity of graph data. For the advent of deep learning, graph anomaly detection with deep learning has received a growing attention recently. In this survey, we aim to provide a systematic and comprehensive review of the contemporary deep learning techniques for graph anomaly detection. We compile open-sourced implementations, public datasets, and commonly-used evaluation metrics to provide affluent resources for future studies. More importantly, we highlight twelve extensive future research directions according to our survey results covering unsolved and emerging research problems and real-world applications. With this survey, our goal is to create a "one-stop-shop" that provides a unified understanding of the problem categories and existing approaches, publicly available hands-on resources, and high-impact open challenges for graph anomaly detection using deep learning.

• The paper presents a detailed taxonomy classifying deep learning methods for detecting anomalies in nodes, edges, sub-graphs, and entire graphs.
• It emphasizes advanced techniques including graph convolutional networks, autoencoders, and reinforcement learning to capture complex relational data.
• It outlines limitations of traditional methods and proposes future research directions for scalability, interpretability, and dynamic graph analysis.

Overview of "A Comprehensive Survey on Graph Anomaly Detection with Deep Learning"

The paper entitled "A Comprehensive Survey on Graph Anomaly Detection with Deep Learning" by Xiaoxiao Ma et al. addresses the evolving landscape of anomaly detection within graph structures through the lens of contemporary deep learning (DL) techniques. Research over the past few decades has illustrated the critical need for effective anomaly detection across a multitude of domains such as security, finance, and medicine, where anomalies can signal detrimental events. Traditional methods in anomaly detection, favoring feature space approaches, fail to capture relational intricacies in datasets, something inherently addressed by graph-based models. Specifically, this paper emphasizes the transition and growing focus on leveraging deep learning techniques to tackle these complex problems of graph anomaly detection.

The core contents of the paper cover an extensive survey of deep learning methodologies applied to graph anomaly detection. With graphs offering a natural representation of relationships among data points—nodes, edges, and sub-graphs—this survey underlines the unique challenges and opportunities posed by graph-structured data. In particular, conventional techniques, often based on linear statistical models or expert-driven heuristics for data format transformation, lack the scalability and adaptability of deep learning approaches, which can learn end-to-end from complex relational data distributions.

Key Contributions and Framework

The paper presents a comprehensive taxonomy that classifies existing work according to the types of anomalous graph objects being targeted: nodes, edges, sub-graphs, and entire graphs. This systematic classification offers researchers clarity on which deep learning frameworks are tailored to specific types of graph anomaly detection, elucidating the strengths and technical nuances of each category.

1. Anomalous Node Detection (ANOS ND): This section explores methods focused on identifying individual nodes whose characteristics (structural or feature-based) differ significantly from the rest. Techniques highlighted include variants of autoencoders, graph convolutional networks (GCNs), and reinforcement learning models that exploit dynamic node interactions over time.
2. Anomalous Edge Detection (ANOS ED): Focused on detecting suspicious or unexpected links within the graph, this section emphasizes approaches like network representation techniques and distribution modeling through autoencoders, providing interpretability and robustness in detection.
3. Anomalous Sub-graph Detection (ANOS SGD): Techniques for detecting suspiciously dense or structurally abnormal sub-graphs are discussed, including methods leveraging bipartite graph modeling and density-based region detection.
4. Anomalous Graph Detection (ANOS GD): This segment covers the detection of entire graphs that deviate from normative graph databases. It reviews graph-level analysis through GNN embeddings, stressing the importance of one-class classification in capture deviations in graph attributes and structure.

Implications and Future Research Directions

The survey highlights several unresolved challenges and proposes future directions, emphasizing the need for more scalable, interpretable, and comprehensive deep learning models targeted at diverse graph data. Among the 12 future directions suggested, salient points include the development of techniques for multi-task learning across graph mining tasks, addressing scalability challenges of deep learning models to handle large-scale graphs efficiently, and enhancing anomaly detection within heterogeneous and dynamic graph environments that mirror real-world data more closely.

Furthermore, incorporating techniques to handle imbalances in graph anomaly detection datasets and providing systematic benchmarking of methods are vital for advancing research in this field. By compiling a rich set of resources, including open-source implementations and benchmark datasets, the paper seeks to facilitate future research and practical applications of graph anomaly detection.

This survey serves as a valuable reference point for researchers and practitioners aiming to leverage the power of deep learning in detecting complex graph anomalies, guiding future research to address the myriad of challenges present in graph-structured data.