Učení reprezentace grafů pro detekci anomálií v grafech

Abstract

The aim of this bachelor thesis is the design and implementation of a modular computational pipeline for anomaly detection in graphs. The proposed system progressively transforms graph structures into node embedding representations, then into distance spaces, and finally into anomaly scoring values used to identify anomalous nodes. The implemented pipeline integrates several types of embedding methods, distance computation techniques, scoring algorithms, and anomaly detection approaches within a unified experimental framework that enables their combination and comparison. The experimental part of the thesis focuses on analysing the influence of individual components and their parameters on anomaly detection performance.