Who Is the Abnormal User: Anomaly Detection Framework based on the Graph Convolutional Networks

• Published in: Web and Big Data pp. 88 - 102
• Main Authors: Zheng, Zetao, Zhu, Jia, Tang, Yong, Du, Jiabing
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Anomaly detection; Graph Convolutional Networks; Graph theory
• ISBN: 9783030260743; 3030260747
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-26075-0_7

Anomaly detection is the identification of items, events or observations which do not conform to an expected pattern in a dataset. It is applicable in a variety of domains, such as intrusion detection, fault detection, medical and public health anomaly monitoring. Existing model usually detects the anomaly according to the data’s feature. However, two drawbacks exist if the model only detects anomaly by using the feature. On the one hand, model could not make use of the relationship between users, which contains a large amount of potential information that can strengthen the prediction ability of the model. On the other hand, existing model could not adjust their learning ability automatically with the increasing of the data. To address the issues referred above, we focus on proposing an anomaly detection system based on the Graph Convolutional Networks (GCN). The framework consists of four mechanisms. It can detect the anomalies by using the user features as well as the relationship between users. Experiment result shows that our framework has outstanding performance compared with other state-of-the-art detection models.