Automated Process Model Extraction from Business Process Descriptions in Natural Language

 

Student: Jan Wieczorek

Title: Automated Process Model Extraction from Business Process Descriptions in Natural Language

Supervisor: Dr. Sebastiaan J. van Zelst

1st Examiner: Prof. Wil van der Aalst

2nd Examiner: Prof. Hermann Ney

Summary

Business processes occur in every organization and have a significant influence on the success of an organization. A process model of a business process allows us to get a schematic overview of the process. Furthermore, the analysis of process models supports the calculation of optimization opportunities for business processes. The generation of process models is very time-consuming and multiple actors are typically involved. Often, most of the information about a process is stored in company-internal documents in the form of textual descriptions. Therefore, in this thesis, we propose an automatic extraction approach of process models from these textual descriptions. Our algorithm consists of two processing steps. The first step is to translate the process description into an intermediate data structure that collects all relevant information. In the second step, the data structure is transformed into a process model. We improve existing approaches by adding further filtering of irrelevant activities and defining more rules for the extraction of the control flow. We implemented our automatic extraction approach and tested it with a set of 44 process descriptions.