Enhancing Class Diagram Dynamics: A Natural Language Approach with ChatGPT
Djaber Rouabhia, Ismail Hadjadj
TL;DR
The paper tackles the challenge of making UML class diagrams dynamic and behaviorally rich by applying ChatGPT-driven NLP to extract methods from natural-language use-case tables for a Waste Recycling Platform. It uses PlantUML to iteratively update static diagrams, validating enhancements against 23 detailed use cases and demonstrating improved accuracy, completeness, and agile-friendliness. Key contributions include a feasible AI-assisted workflow for automatic method identification and integration, a mapped set of added methods across core classes, and a demonstration of how dynamic diagram enrichment can align with rapid development cycles. The work underscores potential benefits for software modeling, while noting limitations related to input quality, reliance on a single use case, and the need for human oversight, suggesting avenues for broader validation and extension to other modeling contexts.
Abstract
Integrating artificial intelligence (AI) into software engineering can transform traditional practices by enhancing efficiency, accuracy, and innovation. This study explores using ChatGPT, an advanced AI language model, to enhance UML class diagrams dynamically, an underexplored area. Traditionally, creating and maintaining class diagrams are manual, time-consuming, and error-prone processes. This research leverages natural language processing (NLP) techniques to automate the extraction of methods and interactions from detailed use case tables and integrate them into class diagrams. The methodology involves several steps: (1) developing detailed natural language use case tables by master's degree students for a "Waste Recycling Platform," (2) creating an initial static class diagram based on these tables, (3) iteratively enriching the class diagram through ChatGPT integration to analyze use cases and suggest methods, (4) reviewing and incorporating these methods into the class diagram, and (5) dynamically updating the PlantUML \cite{plantuml} class diagram, followed by evaluation and refinement. Findings indicate that the AI-driven approach significantly improves the accuracy and completeness of the class diagram. Additionally, dynamic enhancement aligns well with Agile development practices, facilitating rapid iterations and continuous improvement. Key contributions include demonstrating the feasibility and benefits of integrating AI into software modeling tasks, providing a comprehensive representation of system behaviors and interactions, and highlighting AI's potential to streamline and improve existing software engineering processes. Future research should address identified limitations and explore AI applications in other software models.