ארנון שטורם

User stories are a common notation for expressing requirements, especially in agile development projects. While user stories provide a detailed account of the functional requirements, they fail to deliver a holistic view of the domain. As such, they can be complemented with domain models that not only help gain this comprehensive view, but also serve as a basis for model-driven development. We focus on the task of recommending relationships between entities in a domain model, assuming that these entities were previously extracted from a user story collection either manually or through an automated tool. We investigate whether an approach based on supervised machine learning can recommend essential relationships in a domain model more accurately than state-of-the-art rule-based methods. Based on a collection of datasets that we manually labeled and a set of 32 features we engineered, we train a machine learning model by using a random forest classifier. The results indicate that our approach has higher precision and F₁-score than the baseline rule-based methods. Our findings provide preliminary evidence of the suitability of using machine learning to support the development of domain models, especially in recommending relationships between related entities.