MaD: A Dataset for Interview-based BPM in Business Process Management

Business process management focuses on the automatic discovery and optimisation of business process models for a wide range of business scenarios. At the same time, the development of natural language processing (NLP), in particular some large-scale pre-trained language models such as BERT and GPT, has recently achieved great success and become a milestone in many practical fields. We thus propose a new paradigm to automate business process model discovery directly from interview-based natural language documents by applying NLP technologies to the business process modeling area. To train the language models for the business process management domain, we create the Business Process Model and Textual Description (MaD) dataset, which contains 15 business categories and a total of 30,000 BPM-description pairs. Furthermore, we define the automatic as well as human-involved metrics to evaluate the quality of the MaD dataset. The experiment results show that the generated dataset is of high-quality, and suitable for high-level people with professional skills to read and understand. The dataset is available online(1).