In my doctoral thesis, I’m exploring different methods for an automatic extraction and representation of valuable information from data. In particular, I’m working with discrete time series data in which the events of interest occur only very rarely in the form of anomalies. Such anomalies are not necessarily atomic but distributed across several measurements in the temporal stream. A key step here is to develop models for both the evolution of legitimate events and the anomalies by using supervised and unsupervised learning methods. From these models I aim to extract concise representations that summarize the characteristics of temporal patterns, such that, an inexpensive comparison of these representations reveals the commonalities and differences of the original temporal patterns. I also investigate means to exploit and integrate knowledge from Linked Open Data for the purpose of enriching the data stream with additional information.