Year: 2026
Type: article
Abstract: LLMs have become increasingly popular tools for assisting software engineering tasks and specifically with automating the creation of UML design artefacts from requirements specified in a natural language. This work reports on the effectiveness of using LLM technologies to perform the translation of UML use-case diagrams into UML class Diagrams. Such task is traditionally performed by a human and requires both domain understanding and modeling expertise, however, it can be a tedious and labor-intensive manual task prone to errors during its process. Automated creation of UML class Diagrams from UML use-case diagrams requires LLMs to identify the domain entities, the responsibilities of each of those domain entities, and how domain entities relate to each other and therefore creating unique classes. The evaluation part of the work uses several use-case diagrams of varying levels of complexity to complete this task. The results showed that LLMs can effectively identify the crucial domain entities and provide class structures, and to alleviate the modelling efforts. However, there are still issues with the interpretation of ambiguous requirements and the maintenance of their associated semantics. Overall, LLM technologies are an effective aid to the early-stage development of UML class Diagrams; however, the final quality of the UML models is very dependent upon the domain experts that provide further guidance to the LLMs.
Authors: Martina Basic, Marko Vujasinovic
Institution: Veleučilište Aspira
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7128375003