Automated, interactive, and traceable domain modelling empowered by artificial intelligence

Model-Based Software Engineering provides various modelling formalisms for capturing the structural, behavioral, configuration, and intentional aspects of software systems. One of the most widely used kinds of models—domain models—are used during requirements analysis or the early stages of design to capture the domain concepts and relationships in the form of class diagrams. Modellers perform domain modelling to transform the problem descriptions that express informal requirements in natural language to domain models, which are more concise and analyzable. However, this manual practice of domain modelling is laborious and time-consuming. Existing approaches, which aim to assist modellers by automating or semi-automating the construction of domain models from problem descriptions, fail to address three non-trivial aspects of automated domain modelling. First, automatically extracted domain models from existing approaches are not accurate enough to be used directly or with minor modifications for software development or teaching purposes. Second, existing approaches do not support modeller-system interactions beyond providing recommendations. Finally, existing approaches do not facilitate the modellers to learn the rationale behind the modelling decisions taken by an extractor system. Therefore, in this paper, we extend our previous work to facilitate bot-modeller interactions. We propose an algorithm to discover alternative configurations during bot-modeller interactions. Our bot uses this algorithm to find alternative configurations and then present these configurations in the form of suggestions to modellers. Our bot then updates the domain model in response to the acceptance of these suggestions by a modeller. Furthermore, we evaluate the bot for its effectiveness and performance for the test problem descriptions. Our bot achieves median F1 scores of 86%, 91%, and 90% in the Found Configurations, Offered Suggestions, and Updated Domain Models categories, respectively. We also show that the median time taken by our bot to find alternative configurations is 55.5ms for the problem descriptions which are similar to the test problem descriptions in terms of model size and complexity. Finally, we conduct a pilot user study to assess the benefits and limitations of our bot and present the lessons learned from our study in preparation for a large-scale user study.