Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EAGLE: Contrastive Learning for Efficient Graph Anomaly Detection

Jing Ren, Mingliang Hou, Zhixuan Liu, Xiaomei Bai

TL;DR

The paper tackles efficient anomaly detection on heterogeneous graphs under limited supervision. It introduces EAGLE, a self-supervised framework that fuses a graph autoencoder with a discriminator and meta-path–level contrastive learning, enabling robust node representations and anomaly scoring. By sampling positive and negative instance pairs on meta paths and integrating reconstruction and contrastive objectives, EAGLE achieves state-of-the-art performance on three real-world datasets, with noticeable efficiency gains suitable for embedded deployment. The work demonstrates the value of meta-path–driven contrastive learning for preserving heterogeneous semantics while enabling scalable anomaly detection in practice.

Abstract

Graph anomaly detection is a popular and vital task in various real-world scenarios, which has been studied for several decades. Recently, many studies extending deep learning-based methods have shown preferable performance on graph anomaly detection. However, existing methods are lack of efficiency that is definitely necessary for embedded devices. Towards this end, we propose an Efficient Anomaly detection model on heterogeneous Graphs via contrastive LEarning (EAGLE) by contrasting abnormal nodes with normal ones in terms of their distances to the local context. The proposed method first samples instance pairs on meta path-level for contrastive learning. Then, a graph autoencoder-based model is applied to learn informative node embeddings in an unsupervised way, which will be further combined with the discriminator to predict the anomaly scores of nodes. Experimental results show that EAGLE outperforms the state-of-the-art methods on three heterogeneous network datasets.

EAGLE: Contrastive Learning for Efficient Graph Anomaly Detection

TL;DR

The paper tackles efficient anomaly detection on heterogeneous graphs under limited supervision. It introduces EAGLE, a self-supervised framework that fuses a graph autoencoder with a discriminator and meta-path–level contrastive learning, enabling robust node representations and anomaly scoring. By sampling positive and negative instance pairs on meta paths and integrating reconstruction and contrastive objectives, EAGLE achieves state-of-the-art performance on three real-world datasets, with noticeable efficiency gains suitable for embedded deployment. The work demonstrates the value of meta-path–driven contrastive learning for preserving heterogeneous semantics while enabling scalable anomaly detection in practice.

Abstract

Graph anomaly detection is a popular and vital task in various real-world scenarios, which has been studied for several decades. Recently, many studies extending deep learning-based methods have shown preferable performance on graph anomaly detection. However, existing methods are lack of efficiency that is definitely necessary for embedded devices. Towards this end, we propose an Efficient Anomaly detection model on heterogeneous Graphs via contrastive LEarning (EAGLE) by contrasting abnormal nodes with normal ones in terms of their distances to the local context. The proposed method first samples instance pairs on meta path-level for contrastive learning. Then, a graph autoencoder-based model is applied to learn informative node embeddings in an unsupervised way, which will be further combined with the discriminator to predict the anomaly scores of nodes. Experimental results show that EAGLE outperforms the state-of-the-art methods on three heterogeneous network datasets.
Paper Structure (26 sections, 11 equations, 3 figures, 3 tables)

This paper contains 26 sections, 11 equations, 3 figures, 3 tables.

Table of Contents

  1. Related Work
  2. Graph Anomaly Detection
  3. Contrastive Learning
  4. Preliminaries
  5. The Proposed Model: EAGLE
  6. Instance Pair Sampling
  7. Graph AutoEncoder-based Contrastive Learning Model
  8. Attributed Graph Encoder
  9. Attribute Reconstruction Decoder
  10. Structure Reconstruction Decoder
  11. Discriminator
  12. Objective Function
  13. Anomaly Score Computation
  14. Model Analysis
  15. Experiments
  16. ...and 11 more sections

Figures (3)

  • Figure 1: An example of heterogeneous graph, meta-path, and meta-path neighbors. (a) There are three types of nodes, namely Author, Paper, and Venue, and two types of edge, namely Write and Publish in this heterogeneous graph. (b) shows two kinds of meta paths, i.e., Author-Paper-Author, and Paper-Venue-Paper. (c) lists all neighbors of node $A_2$ according to the meta path A-P-A, noting that node $A_2$ itself is also regarded as one of its neighbors.
  • Figure 2: The conceptual framework of EAGLE is composed of four main components: target node selection, instance sampling, contrastive pre-training for graph representation learning, and anomaly score calculation. In this figure, the adjacency matrix PAP and PVP Adj. derived by Paper-Author-Paper and Paper-Venue-Paper meta-paths and the paper attribute matrix Paper Attr. are input to the graph autoencoder for graph representation learning. The process of instance pair sampling is to select positive meta path and negative meta path based on the situation that whether the target node is located in the meta path. The Discriminator aims to distinguish the negative pair from the positive pairs. Finally, the discrimination score and the reconstruction error of the graph decoder are combined to calculate the anomaly score.
  • Figure 3: The impact of embedding dimension on AUC values.

Theorems & Definitions (4)

  • Definition 2.1: Heterogeneous Graph
  • Definition 2.2: Meta path
  • Definition 2.3: Meta path neighbors
  • Definition 2.4: AD on Heterogeneous Graphs