Abstract
Detection of human behavior in On-line Social Networks (OSNs) has become more and more important for a wide range of applications, such as security, marketing, parent controls and so on, opening a wide range of novel research areas, which have not been fully addressed yet.
In this paper, we present a two-stage method for anomaly detection in humans' behavior while they are using a social network. First, we use Markov chains to automatically learn from the social network graph a number of models of human behaviors (normal behaviors); the second stage applies an activity detection framework based on the concept of possible words to detect all unexplained activities with respect to the normal behaviors. Some preliminary experiments using Facebook data show the approach efficiency and effectiveness.