Master Thesis Fabian Friedrich
Automated Generation of Business Process Models from Natural Language Input
© 2010 Fabian Friedrich
A first step to enable effective business process management (BPM) is the design of appropriate conceptual models. These models are used to describe the roles and responsibilities of the employees in an organizational chart, the structure of data, e.g. in an UML class-diagram, and the floow within a process, e.g. as a BPMN-model. They provide the foundation for BPM-initiatives to increase operational performance, but apart from their importance it is also a time and resource intensive task to create such models. On the other hand, information on all the aforementioned subjects is usually available in a company within unstructured textual documents. To reduce the modeling efforts and to accelerate the realization of benefits from a BPM-initiative, we propose an approach for the automated generation of business process models from text documents. In order to create this approach, we analyzed several texts and the corresponding manually created models and derived transformation heuristics from the identified syntactic and semantic patterns. Furthermore, the approach was implemented in a research prototype and evaluated using a similarity metric based on the graph edit distance. Our evaluation has shown encouraging results. In average we were able to generate models which are 76% similar to those created manually by a human.