Analysis of multi-agent interactions with process mining techniques

Process mining and multi-agent models are powerful techniques for the analysis of processes and organizations. However, the integration of both fields has seldom been considered due to the lack of common conceptual background. We propose to close this gap by using Petri nets as an operational semantics and consider process mining a useful addition to monitor and debug multi-agent systems in the development phase. Mining results can be represented in the formalized form of Petri nets that allows to validate or verify the actual behavior. On our way to mining complex interactions within (simulated) organizations, we present a plug-in extension of our Petri net-based agent platform MULAN/CAPA for recording interaction logs. Using process mining, the logs can be mapped by some intermediate steps to agent protocols e.g. represented as Agent UML interaction protocol diagrams. These diagrams are a descriptive representation form that combines organizational and control flow information. Furthermore, they can be mapped to executable Petri net, thus allowing to feed mining results back into the design phase.