Node And Edge Anomaly Detection in Social Networks Using SVM-Clustering and Graph Neural Network Models

Abstract

Anomaly detection is an essential subject with numerous applications and has thus been investigated for decades in numerous research fields, including cyber security, finance, healthcare, social networks, etc. In recent years, numerous strategies have been utilized to detect anomalies in unstructured or multidimensional data. Graphs have the expressive capacity to model data from several domains, including social networks. Literature has a significant amount of research on finding anomalies utilizing structural attributes, ego nets for subspace selection/community analysis. These learning techniques are called shallow mechanisms because they do not completely understand graph structures or patterns. This technique cannot capture the intricate interaction between network nodes and other information modalities. Therefore, the capability of deep learning to solve the challenge of detecting anomalies in social networks was utilized. This article provides a methodology for detecting anomalies in graph data, such as social networks, using graph neural networks' robust representation capability. This paper proposed a model for identifying anomalous connections between social network nodes and edges using deep learning techniques. Clustering and traditional One-class SVM graph neural networks for node anomaly identification was used, and tuned the graph neural networks for edge anomaly detection. The proposed framework produces superior outcomes in comparison to previous baseline models.