Discovering Business Processes in CRM Systems by Leveraging Unstructured Text Data

Recent research has proven the feasibility of using Process Mining algorithms to discover business processes from event logs of structured data. However, many IT systems also store a considerable amount of unstructured data. Customer Relationship Management (CRM) Systems typically store information about interactions with customers, such as emails, phone calls, meetings, etc. These activities are characteristically made up of unstructured data, such as a free text subject and description of the interaction, but only limited structured data is available to classify them. This poses a problem to the traditional Process Mining approach that relies on an event log made up of clearly categorised activities. This paper proposes an original framework to mine processes from CRM data, by leveraging the unstructured part of the data. This method uses Latent Dirichlet Allocation (LDA), an unsupervised machine learning technique, to automatically detect and assign labels to activities. This framework does not require any human intervention. A case study with real-world CRM data validates the feasibility of our approach.