Mining Probabilistic Declarative Process Models

The management of business processes has recently received a lot of attention from companies, since it can support efficiency improvement. We present an approach for mining process models that first induces a model in the SCIFF logical language and then translates the model into Markov logic, a language belonging to the field of statistical relational learning. Markov logic attaches weights to first-order contraints, in order to obtain a final probabilistic classification of process traces better than the purely logical one. The data used for learning and testing belong to a real database of university students' careers.