LLM-Generated Software Requirements from GitHub Issues

This dataset contains software requirements automatically generated from bug reports and feature requests extracted from the three most popular machine learning repositories on GitHub: Scikit-learn, TensorFlow, and Transformers. The dataset is structured into issue data, generated requirements, and evaluations based on three well-defined criteria.

Dataset Structure

• issues.csv: Contains issue titles along with their corresponding repository names and unique identifiers.
• Requirements Files: These files store the requirements generated by LLMs for each issue, categorized by different prompting methods:
  • few_shot_requirements.csv
  • zero_shot_requirements.csv
  • expert_requirements.csv
  • expert_few_shot_requirements.csv
• Evaluation Files: These files contain the assessment of the generated requirements based on three key quality criteria: Unambiguity, Understandability, and Singularity. The evaluations are also divided by prompting methods:
  • few_shot_evaluation.csv
  • zero_shot_evaluation.csv
  • expert_evaluation.csv
  • expert_few_shot_evaluation.csv