Using machine learning to enhance automated requirements model transformation

Textual specification documents do not represent a suitable starting point for software development. This issue is due to the inherent problems of natural language such as ambiguity, impreciseness and incompleteness. In order to overcome these shortcomings, experts derive analysis models such as requirements models. However, these models are difficult and costly to create manually. Furthermore, the level of abstraction of the models is too low, thus hindering the automated transformation process. We propose a novel approach which uses high abstraction requirements models in the form of Object System Models (OSMs) as targets for the transformation of natural language specifications in conjunction with appropriate text mining and machine learning techniques. OSMs allow the interpretation of the textual specification based on a small set of facts and provide structural and behavioral information. This approach will allow both (1) the enhancement of minimal specifications, and in the case of comprehensive specifications (2) the determination of the most suitable structure of reusable requirements.