Abstract
Process discovery techniques focus on learning a process model starting from a given set of logged traces. The majority of the discovery approaches, however, only consider one set of examples to learn from, i.e., the log itself. Some recent works on declarative process discovery, instead, advocated the usefulness of taking into account two different sets of traces (a.k.a. positive and negative examples), with the goal of learning a set of constraints that is able to discriminate which trace belongs to which set. Sometimes, however, too many possible sets of constraints might be available, thus nullifying the discovery effort. Therefore, some preference criteria would be helpful to guide the discovery process towards a set of constraints among the many. In this work, we present an approach for the discovery of declarative models providing the possibility, from the user viewpoint, of specifying preferences on activities and constraint templates to be used to build the final set of constraints. Such preferences are used to guide the discovery process, so that the output set will include, if possible, the preferred constraints, thus exploiting some expert knowledge about the desired outcome. The approach is grounded in a logic-based framework that provides a sound and formal meaning to the notion of expert preferences.