On the mining of artful processes

Artful processes are those processes in which the experience, intuition, and knowledge of the actors are the key factors in determining the decision making. These knowledge-intensive processes are typically carried out by the “knowledge workers”, such as professors, managers, researchers. They are often scarcely formalised or completely unknown a priori, and depend on the skills, experience, and judgment of the primary actors. Artful processes have goals and methods that change quickly over time, making them difficult to codify in the context of an enterprise application. Knowledge workers cannot be realistically expected to instruct the assistive system by modelling their artful processes: it would be time-consuming both in the initial definition and in the potential continuous revisions. To make things worse, time is the crucial resource that usually knowledge workers indeed lack. Despite the advent of structured case management tools, many enterprise processes are still “run” over emails. Thus, reverse engineering workflows of such processes and their integration with artefacts and other structured processes can accurately depict the enterprise’s process landscape. A system able to infer the models of the processes laying behind the email messages exchanged would be valuable and the result could materialise almost freely. This is the purpose of our approach, which is the core of this thesis and is named MailOfMine. Its investigation mainly resides in the Machine Learning area. More specifically, it relates to Information Retrieval (IR) and Process Mining (PM). We adopted well-known IR techniques in order to extract the activities out of the email messages. We propose a new algorithm for PM in order to discover the temporal rules that the activities adhere to: MINERful. The set of such rules, intended as temporal constraints, constitute the so called declarative modelling of workflows. Declarative models differ from the imperative in that they do not explicitly represent every possible execution that a process can be enacted through, i.e., there is no graph-like structure determining the whole evolution of a process instance, from the beginning to the end. They establish a set of constraints that must hold true, whatever the evolution of the process instance will be. What is not explicitly declared to be respected, is allowed. The reader can easily see that it is better suited to processes subject to frequent changes, with respect to the classical approach. From a more abstract perspective, this work challenges the problem of discovering highly flexible workflows (such as artful processes), out of semi-structured information (such as email messages).