Exploring the Use of Large Language Models in Requirements Engineering Education: An Experience Report with ChatGPT 3.5

• Savio Sousa Sampaio UFAM
• Márcia Sampaio Lima UEA
• Eriky Rodrigues de Souza UFAM
• Maria Alcimar Meireles UFMA
• Marcela Savia Pessoa UEA
• Tayana Uchoa Conte UFMA

Resumo

Large Language Models (LLMs) are becoming common in educational settings. This trend presents a challenge for teachers, who must focus on teaching the proper usage of LLMs. In the context of Software Engineering (SE), ChatGPT can support various software development tasks. This work reports an experience with students using ChatGPT 3.5 to support the Requirements Engineering (RE) phase. We conducted a two-phase study with 42 students. First, the students elicited requirements for systems using RE techniques. Then, the students used ChatGPT 3.5 to generate requirements for the same systems. Finally, they compared both sets of requirements based on equivalence, innovation, and relevance. On average, 65.26% of the requirements generated by ChatGPT were considered equivalents to the requirements the students had elicited. However, students reported that ChatGPT generates broad and non-specific requirements. Students also reported that ChatGPT 3.5 can foster the requirements elicitation, but it is necessary to establish well-defined prompts for generating requirements.

Palavras-chave: Requirement Elicitation, ChatGPT 3.5, Software engineering education