Situation Mining: Event Pattern Mining for Situation Model Induction

Computational situation assessment (SA) systems support human control center operators in situation monitoring, i.e., detecting and tracking relevant object and event constellations in their observed environment. SA systems frequently employ deductive reasoning techniques implemented in Complex Event Processing or rule engines to solve this real-time pattern recognition problem, by matching data sensed from the monitored environment against templates for those situations, characterizing the event patterns of interest. Hence, they require explicitly formalizing the sought-after types of situations, demanding human domain experts to conceptually model their cognitive situation hypotheses, which represents a time-consuming and non-trivial task. To overcome this situation knowledge acquisition bottleneck, we therefore propose an approach for inductive situation modeling to leverage existing data sets of recorded situations: We contribute a dedicated situation mining algorithm, which bootstraps situation model acquisition by automatically mining behavioral models of situations, so-called situation evolution models, from already observed situation instances. The feasibility of our approach is examined on a case study from the domain of road traffic incident management, to demonstrate how it turns previously implicit knowledge hidden in the situation instances into explicit situation knowledge that can be inspected and queried for situation analytics, and sketch how the derived situation evolution models can be used within a Model-Driven Engineering framework to directly generate the corresponding rule code for automated situation assessment.