A Comparison of Transforming the User Stories and Functional Requirements into UML Use Case Diagram

Abstract

Software development life cycle is a continuous process for every developer community, including establishing the user requirements and system design until software maintenance. The system design phase contains the transformation of all-natural language requirements into models. The Unified Modelling Language (UML) use case model is regarded as one of the most helpful diagrams for a software developer to understand the higher version of the software blueprint. It’s very useful to reflects the business requirements for any proposed software system. Manual analysis requires significant time and effort, highlighting the need for automated assistance. This study aims to enhance the machine learning techniques that have been utilized to transform user stories and functional requirements into a UML use case diagram. This research also investigate the use of an automated Natural Language Processing (NLP) tool and framework for mapping use case diagram elements from user stories and functional requirements written in natural languages. The proposed approach using a NLP based machine learning framework called BERT to predict and extract actors, use cases, and the relationship between actors and use cases. This approach has been applicable for both user stories and functional requirements. A dataset has been pre-processed in order to train this model and collected from various sources like Kaggle, software projects. There are 11 user stories and 11 functional requirements were selected to test the proposed solution. The data set from benchmark repositories. The extracted results during the whole NLP process are then mapped into actors, use cases, and relationships. This study was evaluated using both accuracy and precision values for performance measures. By applying NLP technologies, this research offers complete support for extracting elements and generating a use case diagram and revealed the 88% accuracy and 88% precision.