Process Mining meets argumentation: Explainable interpretations of low-level event logs via abstract argumentation

Abstract

Generally, companies and organizations can greatly improve their business processes by suitably monitoring and analyzing the log data that they gather for these processes in the form of traces. We here consider the challenging scenario where there is an abstraction gap between the “low-level” events composing the traces and the “high-level” activities on which analysts are used to reason. Herein, we aim at supporting the analysis of an ongoing process instance w by addressing the online interpretation problem of translating the event that has just been generated within w into its “high-level” meaning, i.e. into the step of the activity instance it corresponds to. We model this interpretation problem as a dispute through an Abstract Argumentation Framework (AAF), so that the computations of valid interpretations and explanations of the invalidity of other interpretations are elegantly translated into instances of the classical AAF acceptance problem. A thorough empirical analysis is performed to assess the effectiveness and efficiency of the proposal, against both synthesized and real log data.