Modern software engineering faces complex challenges, particularly in accurately and efficiently modeling systems from textual specifications. UML class diagrams are essential tools for representing the static structure of a software system, but creating them manually can be laborious and error-prone. The emergence of deep learning techniques, more specifically Large Language Models (LLMs) opens new perspectives for automatically extracting these UML diagrams from textual descriptions. This article explores the use of deep learning, particularly large language models, to automate this extraction process. We review existing approaches, including linguistic and statistical methods, and propose a new method based on deep learning models to improve accuracy and flexibility compared to current extraction tools.