Extracting Domain Models from Textual Requirements in the Era of Large Language Models

Abstract

Requirements Engineering is a critical part of the software lifecycle, describing what a given piece of software will do (functional) and how it will do it (non-functional). Requirements documents are often textual, and it is up to software engineers to extract the relevant domain models from the text, which is an error-prone and time-consuming task. Considering the recent attention gained by Large Language Models (LLMs), we explored how they could support this task. This paper investigates how such models can be used to extract domain models from agile product backlogs and compare them to (i) a state-of-practice tool as well as (ii) a dedicated Natural Language Processing (NLP) approach, on top of a reference dataset of 22 products and 1,679 user stories. Based on these results, this paper is a first step towards using LLMs and/or tailored NLP to support automated requirements engineering thanks to model extraction using artificial intelligence.