Extracting UML class diagrams from software requirements in Thai using NLP

Abstract

In software development, requirements, normally written in natural language, are documents that specify what users want in software products. Software developers then analyze these requirements to create domain models represented in UML diagrams in an attempt to comprehend what users need in the software products. These domain models are usually converted into design models and finally carried over into classes in source code. Thus, domain models have an impact on the final software products. However, creating correct domain models can be difficult when software developers are not skilled. Moreover, even for skilled developers, when requirements are large, wading through all requirements to create domain models can take times and might result in errors. Therefore, researchers have studied various approaches to apply natural language processing techniques to transform requirements written in natural language into UML diagrams. Those researches focus on requirements written in English. This paper proposes an approach to process requirements written in Thai to extract UML class diagrams using natural language processing techniques. The UML class diagram extraction is based on transformation rules that identify classes and attributes from requirements. The results are evaluated with recall and precision using truth values created by humans. Future works include identifying operations and relationships from requirements to complete class diagram extraction. Our research should benefit Thai software developers by reducing time in requirement analysis and also helping novice software developers to create correct domain models represented in UML class diagram.