Security monitoring tools, such as honeypots, IDS, behavioral analysis or anomaly detection systems, generate large amounts of security events or alerts. These alerts are often shared within some communities using various alert sharing systems. Our research is focused on analysis of the huge amount of data present in these systems. In this work we focus on summarizing all alerts and other information known about a network entity into a measure called reputation score expressing the level of threat the entity poses. Computation of the reputation score is based on estimating probability of future attacks caused by the entity.