An Anomaly Detection Framework for Propagation Networks Leveraging Deep Learning

In the Internet era, communication networks are endless, and all kinds of communication anomalies are also dazzling, so it becomes especially important to detect anomalies in social network communication. Graph networks with their excellent ability to capture spatial information have been well applied in the propagation field. In this paper, we develop a graph convolutional deep self-coding framework AD-GCN, which is the first learning algorithm used for anomaly detection in social network propagation. It is based on a deep self-coder, on top of which the popular Graph Convolutional Network (GCN) is introduced to learn the underlying distribution of graph structures and node attributes and to construct the reconstruction error and detect anomalies from both structure and loss perspectives. We have collected a real dataset of hot topics in microblogging and tested accordingly on this dataset to achieve better results.