A Sketch of a Deep Learning Approach for Discovering UML Class Diagrams from System’s Textual Specification
Yves Rigou, Dany Lamontagne, Ismaïl Khriss · 2020 1st International Conference on Innovative Research in Applied Science, Engineering and Technology (IRASET) · 2020
Drafting a formal or semi-formal model describing the functional requirements of a system from a textual specification is a prerequisite in the context of a model-driven engineering approach, such as the model-driven architecture initiative proposed by OMG. This model, called a platform-independent model (PIM), is used to derive automatically or semi-automatically the source code of a system. Different knowledge-based approaches have been proposed to extract a PIM from a textual specification automatically. These approaches use a predefined set of rules to perform this discovery. These approaches impose several restrictions on the way a specification is written. The emergence of machine learning techniques and more specifically of deep learning and their obvious success among others in several tasks in automatic language processing, such as speech recognition and translation, suggests the possibility of using these techniques to reach our objective. In this paper, we review state of the art in the domain and we sketch a rough deep learning approach to achieve our objective of extracting a PIM from the textual specification of a system.