Extracting Formal Process Model Information from Natural Language Text Descriptions using Machine Learning

Business process management helps businesses and organizations with understanding, improving, and automating reoccurring workflows, called business processes. One integral part to this is a formal business process model, which is expensive to create, as it needs collaboration between process participants and process analysts. Automatic generation of formal business process models is a research field working on facilitating the creation of formal process models, by extracting the information contained in existing natural language process descriptions and synthesizing a formal process model. While this field of research has become increasingly popular in recent years, many process information extraction approaches still rely on hand-written rules, which makes them hard to transfer to new application domains, description languages, or process modeling languages. Approaches based on machine learning and deep learning would provide significant benefits in this regard, as extraction rules are derived from data automatically and are therefore easier to adjust. Yet, these approaches are still rare, which can be explained by their need for large amounts of training data. This thesis presents three options to mitigate the lack of large training datasets in process information extraction, which are as follows. Facilitating the collection of new training data via proper tool support, more efficient use of existing training data, and the application of pretrained language models. Thus, this thesis paves the way for successful applications of text-to-model methods in the future.