Emails analysis for business process discovery

Most often, process mining consists of discovering models of actual processes from structured event logs. However, some business processes (BP), or at least some parts of them, are not necessary supported by an information system (IS), and consequently do not leave any structured events log. Therefore, applying traditional process mining techniques would generate a partial view of such processes. Process actors often rely on communication tools to collaboratively execute their business processes in such situations. However, given the unstructured nature of communication tools traces, process mining techniques could not be applied directly; thus it is necessary to generate structured event logs by recognizing the process-related items (activities, actors, instances, etc.). In this paper, we address this challenge in order to mine business processes from email exchange traces. We introduce an approach that minimizes users' efforts to manage the growing amounts of exchanged emails: It enables to collaboratively, and gradually build an annotated corpus of messages, and to automatically classify these ones into process, instance and activity IDs using machine learning techniques. Compared to related works, we facilitate the task of obtaining annotated datasets and we investigate the use of email exchange histories, correspondent, references and named entities for building clustering and classification features. The proposed approach is evaluated through a proof of concept and successfully experimented on an email dataset.