Early Results of an AI Multiagent System for Requirements Elicitation and Analysis

Abstract

In agile software development, user stories capture requirements from the user’s perspective, emphasizing their needs and each feature’s value. Writing concise and quality user stories is necessary for guiding software development. Alongside user story generation, prioritizing these requirements ensures that the most important features are developed first, maximizing project value. This study explores the use of Large Language Models (LLMs) to automate the process of user story generation, quality assessment, and prioritization. We implemented a multi-agent system using Generative Pre-trained Transformers (GPT), specifically GPT-3.5 and GPT-4o, to generate and prioritize user stories from the initial project description. Our experiments on a real-world project demonstrate that GPT-3.5 handled user story generation well, achieving a higher semantic similarity score comnpared to the GPT-4o. Both models showed consistent performance in prioritizing requirements, effectively identifying the core features of the application. These early results indicate that LLMs have significant potential for automating requirements analysis, particularly generating and prioritizing user stories.