Discovering protocols and organizational structures in workflows

Workflow mining consists in analyzing the execution traces of a collaborative system in order to create a workflow. Most of the works in this area focus on the process discovery from log files, while very few of them address the organization aspect. On the other hand, the Agent technology, suitable for the design and development of cooperative systems, offers interaction languages and high-level protocols allowing the representation of fine-grained organizational structures (hierarchy, federation, coalition, market, among others). In this paper, we show how to enrich log files in order to integrate conversations among the actors of processes using a performative-based agent communication language (FIPA-ACL). This enrichment allows us to define algorithms to discover protocols and organizational structures. The organizational structures are represented by graphs where the nodes are the actors and the arcs relations among them (delegation, cooperation, information exchange, and so on). The interaction protocols (auction, contract net) are described by Petri nets. We provide some discovery algorithms as well as a brief description of a Java implementation coupled with a visualization tool called AGNA.