Abstract:

The emergence of Large Language Models (LLMs) has revolutionized various domains, including education. Within the ever-changing field of computer science, particularly in software engineering education, there is an ongoing demand for innovative pedagogical approaches to ensure that students develop a deep understanding of complex concepts. This research investigates the integration of Natural Language Processing (NLP) techniques, specifically LLMs, to teach the transformation of real-world problems expressed in natural language into standardized technical diagrams utilizing a textual notation that interfaces with visual diagram tools.

Unified Modeling Language (UML) diagrams are highly relevant tools that serve as visual representations of software systems, facilitating the design and communication of software architecture. However, students often face difficulties in comprehending the intricate relationships and structures depicted in UML diagrams. The proposed approach leverages the natural language processing (NLP) capabilities of large language models (LLMs) to guide students through the process of translating real-world scenarios into UML diagrams.

The research methodology involves deploying pre-trained and fine-tuning LLM models for interpreting and generating UML diagrams from natural language descriptions. The LLM-driven system employs state-of-the-art open-source language models, such as Llama2, to process and analyze textual descriptions of UML elements. Through advanced language understanding, the system translates natural language input into corresponding UML constructs, fostering a dynamic and responsive educational interface. A key feature of the proposed approach is the adaptive nature of the learning environment. This tool continuously analyzes student interactions, tailoring feedback and instructional content to address specific learning needs. This adaptability fosters a personalized and effective learning experience, catering to diverse learning styles.

The study will be conducted with a cohort of 25 second-year software engineering students, employing a combination of controlled experiments and qualitative assessments. To evaluate the effectiveness of the LLM-enhanced teaching approach, we will utilize performance metrics, user satisfaction surveys, and pre- and post-instruction evaluations.

Expected outcomes of the research include enhanced comprehension of UML concepts, improved retention, and increased confidence in translating real-world problems into UML diagrams. The research seeks to identify the strengths and limitations of the LLM-driven approach and contribute insights into the potential transformation of software engineering education.

The findings of this research underscore the potential of LLMs to enhance software engineering education by providing a more effective and engaging approach to teaching. The proposed LLM-empowered environment offers a promising alternative to traditional methods, enabling students to develop a deeper understanding of software engineering concepts and their practical application in software design and documentation. Furthermore, the study contributes to the ongoing evolution of pedagogical strategies in software engineering education, fostering a more profound understanding of complex concepts through the integration of cutting-edge technologies.

Keywords:

Generative AI, Large language models, UML diagrams, software engineering education, real-world problems, problem-solving, active learning.