Hyperbolic Graph Embeddings: a Survey and an Evaluation on Anomaly Detection

TL;DR

This paper surveys hyperbolic graph embedding models and evaluates their effectiveness in anomaly detection, showing they outperform Euclidean methods in capturing complex graph structures and providing an open-source library for future research.

Contribution

It offers a comprehensive review of hyperbolic graph embedding models and evaluates their performance on anomaly detection tasks, highlighting their advantages over Euclidean approaches.

Findings

VAE achieved 94% F1-score on Elliptic dataset.

HGCAE scored 80% on Cora dataset.

Hyperbolic methods outperform Euclidean ones in complex anomaly detection.

Abstract

This survey reviews hyperbolic graph embedding models, and evaluate them on anomaly detection, highlighting their advantages over Euclidean methods in capturing complex structures. Evaluating models like \textit{HGCAE}, \textit{\(\mathcal{P}\)-VAE}, and \textit{HGCN} demonstrates high performance, with \textit{\(\mathcal{P}\)-VAE} achieving an F1-score of 94\% on the \textit{Elliptic} dataset and \textit{HGCAE} scoring 80\% on \textit{Cora}. In contrast, Euclidean methods like \textit{DOMINANT} and \textit{GraphSage} struggle with complex data. The study emphasizes the potential of hyperbolic spaces for improving anomaly detection, and provides an open-source library to foster further research in this field.