From Problem Descriptions to User Stories: Utilizing Large Language Models through Prompt Chaining

Abstract

In software development, generating user stories from problem descriptions plays a key role in understanding and implementing user needs. Leveraging Large Language Models (LLMs) for this task holds immense potential in streamlining the process and improving the accuracy of user story extraction. This paper presents a novel approach, utilizing prompt chaining with LLMs to transform problem descriptions into user stories. The primary focus is extracting the 'who,' 'what,' and optional 'why' components of user stories from specifications of the domain written in natural language (NL). The proposed method involves a four-step prompt chaining strategy that includes identifying roles, grouping functionalities based on roles, determining the rationale behind each role's functionalities, and, finally, crafting user stories in the Connextra notation. Evaluation of problem descriptions of five different systems demonstrated the effectiveness of the approach in accurately extracting Roles and Functions of user stories. The results showcased high average precision (94%), recall (8 9%), and F1 scores (91%), indicating a strong alignment between the extracted user stories and the Gold Standard. The results indicate that the proposed method holds significant promise for expediting requirement analysis and software development processes. © 2024 IEEE.