Anomalous change detection in time-evolving OSNs

Recently, the automatic highlighting of anomalous changes in a sequence of social graph snapshots is receiving growing interest due to its numerous applications. For instance, it may be helpful for the identification of attackers or risky users in Online Social Networks (OSNs). Indeed, dynamically monitoring and learning the friendship patterns of users in a large social graph over time for any anomalous change often reflects and predicts significant events or attacker's behaviors. In this paper, we focus on anomalous changes that happen in the neighborhood of OSN users. Our main goal is to detect those users whose changes in the structure of their subgraph deviate from their own previous change patterns and from those of other nearest users in the graph. Our approach returns a list of these users by ranking them based on the value of their change deviation. We analyze the performance of our approach on a real Google+ dataset.