View Item 
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        JavaScript is disabled for your browser. Some features of this site may not work without it.

        Browse

        All of UU Student Theses RepositoryBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

        From Interview to Process Model: Using Large Language Models for Process Discovery

        Thumbnail
        View/Open
        ThomUffing_Final_Thesis.pdf (3.597Mb)
        Publication date
        2025
        Author
        Uffing, Thom
        Metadata
        Show full item record
        Summary
        The discovery of business processes is a crucial step for organisations that wish to understand their operations. This Process Discovery phase is of ten executed by process experts who interview domain experts to gather knowledge about the process. After the interview, the process expert anal yses the collected data and manually transforms this into an intermediate process model that can be used for validation. This manual approach is time-consuming due to the amount of work required to create the process model combined with the planning of multiple (validation) meetings, often with various domain experts. In this thesis, we present a methodology of using Large Language Models to create process models from interview transcripts. Previous approaches focussed on generating process models from process descriptions or log data from information systems, this study differentiates by extracting process knowledge with interview data as input. We evaluate our methodology by comparing the process models gener ated from a Large Language Model with manually created models by Mas ter’s students who completed a Business Process Management course. Our results demonstrate that the models generated from the Large Language Model provide similar results for the extraction of the transcripts and mod elling of actors, however, the generated models contain a lower number and similarity of activities, events, and gateways than the human-generated ones. These findings suggest the value of this methodology for supporting the Pro cess Discovery phase, which can result in time savings for the process- and domain experts
        URI
        https://studenttheses.uu.nl/handle/20.500.12932/48957
        Collections
        • Theses
        Utrecht university logo