Machine learning-based incremental learning in interactive domain modelling

In domain modelling, practitioners manually transform informal requirements written in natural language (problem descriptions) to more concise and analyzable domain models expressed with class diagrams. With automated domain modelling support using existing approaches, manual modifications may still be required in extracted domain models and problem descriptions to make them more accurate and concise. For example, educators teaching software engineering courses at universities usually use an incremental approach to build modelling exercises to restrict students in using intended modelling patterns. These modifications result in the evolution of domain modelling exercises over time. To assist practitioners in this evolution, a synergy between interactive support and automated domain modelling is required. In this paper, we propose a bot-assisted approach to allow practitioners perform domain modelling quickly and interactively. Furthermore, we provide an incremental learning strategy empowered by machine learning to improve the accuracy of the bot's suggestions and extracted domain models by analyzing practitioners' decisions over time. We evaluate the performance of our bot using test problem descriptions which shows that practitioners can expect to get useful support from the bot when applied to exercises of similar size and complexity, with precision, recall, and F2 scores over 85%. Finally, we evaluate our incremental learning strategy where we observe a reduction in the required manual modifications by 70% and an improvement of F2 scores of extracted domain models by 4.2% when using our proposed approach and learning strategy together.