Generated Software Requirements From User Stories

This dataset contains software requirements generated by multiple Large Language Models (LLMs) based on a set of standardized user stories. It was created to support an experimental study examining how different LLMs perform in requirements generation and how structured prompting affects the quality of the produced requirements. For every requirement, the dataset includes the generated text, the source user story, the model used, the prompting condition, and several quality-related measures calculated during the experiment: Classification result (Functional / Non-Functional) Ambiguity Score Semantic Similarity to the user story Passive-voice detection 5W1H completeness score (Who, What, When, Where, Why, How) Readability metrics (Flesch-Kincaid Eease, Flesch-Kincaid Grade Level, Gunning Fog Index) The dataset is intended for research on LLM behavior in requirements generation, comparison of prompting strategies, and analysis of requirement clarity, completeness, and language quality.