Interpretable Graph-Level Anomaly Detection via Contrast with Normal Prototypes

2026-02-11

AI Summary

This work addresses the limited interpretability of existing graph-level anomaly detection methods, which often fail to provide intuitive explanations grounded in normal instances. To overcome this, we propose ProtoGLAD, a novel framework that introduces concrete normal graphs as prototypes—rather than abstract vectors—and performs unsupervised anomaly detection by explicitly comparing a test graph against multiple normal prototype graphs iteratively mined from the data. By integrating point-set kernels with graph distance metrics, ProtoGLAD effectively identifies anomalies while offering human-interpretable explanations. Experimental results demonstrate that ProtoGLAD achieves detection performance on par with state-of-the-art methods across multiple real-world datasets, while significantly enhancing the clarity and intuitiveness of its explanations.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

The task of graph-level anomaly detection (GLAD) is to identify anomalous graphs that deviate significantly from the majority of graphs in a dataset. While deep GLAD methods have shown promising performance, their black-box nature limits their reliability and deployment in real-world applications. Although some recent methods have made attempts to provide explanations for anomaly detection results, they either provide explanations without referencing normal graphs, or rely on abstract latent vectors as prototypes rather than concrete graphs from the dataset. To address these limitations, we propose Prototype-based Graph-Level Anomaly Detection (ProtoGLAD), an interpretable unsupervised framework that provides explanation for each detected anomaly by explicitly contrasting with its nearest normal prototype graph. It employs a point-set kernel to iteratively discover multiple normal prototype graphs and their associated clusters from the dataset, then identifying graphs distant from all discovered normal clusters as anomalies. Extensive experiments on multiple real-world datasets demonstrate that ProtoGLAD achieves competitive anomaly detection performance compared to state-of-the-art GLAD methods while providing better human-interpretable prototype-based explanations.

Problem

• graph-level anomaly detection
• interpretability
• normal prototypes
• anomaly explanation
• unsupervised learning

Innovation

• graph-level anomaly detection
• interpretable AI
• normal prototypes
• point-set kernel
• unsupervised learning