Exploring ChatGPT Prompt Engineering for Business Process Models Semantic Quality Improvement

Business process modeling aims to enhance the comprehension of company processes, enabling decision-makers to attain strategic objectives. Nevertheless, inexperienced systems analysts who lack domain knowledge may produce low-quality models, thereby impacting the overall effectiveness of the modeling process. Large language models are bringing about a revolutionary transformation in the application of knowledge within various domains. This paper presents preliminary experimental findings on the potential application of Large language models in Business Process Modeling semantic quality improvement, specifically focusing on the extent to which these technologies can aid the modeler by suggesting improvements. In our study, we delve into the knowledge of GPT-4, a powerful language model. We aim to investigate its capabilities by employing various combinations of prompts, incorporating proposed textual syntax, and integrating contextual domain knowledge. Our objective is to leverage these approaches to improve the quality of business process models. Our findings indicate that the knowledge generated by GPT-4 is predominantly generic, encompassing ambiguous and general concepts that extend beyond the specific domain. However, when we apply our proposed prompting techniques, we observe a notable improvement in the specificity and comprehensiveness of the results. The utilization of these prompts helps to refine the generated knowledge, leading to more specific and comprehensive outcomes that align closely with the intended domain.