Goal Model Extraction from User Stories Using Large Language Models

In agile software development, goal modeling is vital for understanding the relationships among user stories, commonly used to capture stakeholders’ needs. Manual construction of goal models faces challenges, such as transforming lower-level user stories into higher-level models and capturing implicit goals. This paper presents early research proposing a technique using Large Language Models (LLMs), like GPT-4, to automatically generate goal models from user stories. The approach employs Iterative Prompt Engineering to guide the LLM in extracting intentional elements and generating XML representations using the Goal-oriented Requirements Language (GRL), visualized with the jUCMNav tool. Our primitive qualitative evaluation indicates that GPT-4 can produce GRL models that are acceptable and understandable. Despite the generic nature of LLM-generated models, there is a potential for their use in requirements modeling, particularly in exposing soft goals not immediately apparent to stakeholders new to the domain.