Authors:
Virginia Niculescu
;
Maria-Camelia Chisăliță-Crețu
;
Cristina-Claudia Osman
and
Adrian Sterca
Affiliation:
Babeș-Bolyai University, Cluj-Napoca, Romania
Keyword(s):
Model-Driven Development, Large Language Models, Conceptual Diagrams, Business Process Model and Notation, Entity-Relationship Diagrams, User Productivity, Business Analysts.
Abstract:
The recent rise of Large Language Models (LLMs) suggests the possibility for users with different levels of expertise to generate software applications from high-level specifications such as formatted text, diagrams or natural language. This would enhance productivity and make these activities accessible to users without a technical background. Approaches such as Model-Driven Engineering (MDE) and Workflow Management Systems (WfMSs) are widely used to enhance productivity and streamline software development through automation. This study explores the feasibility of using LLMs, specifically ChatGPT, in software development, focusing on their capability to assist business analysts (BAs) in generating functional applications. The goal of this paper is threefold: (1) to assess the extent to which LLMs comprehend conceptual model diagrams, (2) to evaluate the reliability of diagram-based code generation, and (3) to determine the level of technical knowledge required for users to achieve v
iable solutions. Our methodology evaluates the effectiveness of using LLMs to generate functional applications starting from BPMN process diagrams and Entity-Relationship (ER) diagrams. The findings provide insights into the reliability and limitations of LLMs in diagram-based software generation, the degree of technical expertise required, and the prospects for adopting LLMs as tools for BAs.
(More)