Communication networks are formed by a group of entities exchanging information with each other, such as people's phone calls, email communications or even network traffic. Real world events that change people's communication behaviour will be reflected in the communication networks. Detecting these events is crucial for tasks such as counter terrorism, network surveillance, traffic management and illegitimate use of resources in organizations (e.g. a student starting a commercial operation out of a university network). Content-based detection methods identify events through analyzing the contents of the communications, such as the contents of emails or the conversation of phone calls. However, the contents are not always available for various reasons, such as privacy protection. Link-based detection methods circumvent this problem by exploiting the linkage information of the networks. They model the communication networks as graphs with vertices representing entities and edges communications. By studying the dynamics of activities in communication networks, they can identify those vertices with abnormal variations of activities. In this project, we develop a framework for applying link-based analysis on communication networks. Guidelines are proposed for each step, from feature selection to result evaluation. Several detection methods that can be easily embedded in the framework are also proposed. The effectiveness of the framework and methods is assessed on both network traffic data and email communication networks.
展开▼
