Empirical Data Package for "Quality Assessment of Software Requirements Using Artificial Intelligence Methods: A Systematic Literature Review"

This dataset, referred to as the "Empirical Data Package", contains all systematically extracted data for the following study: E. Wolf, A. Trendowicz, J. Siebert, "Quality Assessment of Software Requirements Using Artificial Intelligence Methods: A Systematic Literature Review".

Research objective:
The study investigates how AI techniques and models are applied to assess and improve the quality of software requirements. It addresses three research questions:
- RQ1: Which AI methods and models are used for automated assessment of requirements quality?
- RQ2: Which quality aspects and associated metrics are considered for assessment?
- RQ3: Which datasets are used to create, evaluate, and adjust requirements quality assessment models?

Dataset contents:
- Screening logs, including title-abstract review, snowballing results, and the final set of 26 primary studies.
- Demographic information such as type of contribution and research context (industry vs. academia).
- Quality assessment data per study, including applied criteria and scoring.
- Extracted mappings for each study: AI methods, evaluation metrics, quality aspects, purposes of AI application, RE phases targeted, and datasets used.
- Organized per research question (RQ1–RQ3) in Excel worksheets with grouping and content tabs.

Notable findings included in the dataset:
- Requirement quality aspects from multiple sources can largely be mapped to the INVEST framework.
- AI models primarily focus on detecting issues; few provide actionable improvements.
- AI methods rarely cover all INVEST aspects, with some quality criteria neglected.
- The field exhibits heterogeneity in datasets, labeling strategies, and evaluation metrics, highlighting challenges for reproducibility, generalization, and adoption in practice.
- Recent AI approaches, including GenAI models, offer opportunities for explanation and recommendations, but require careful adaptation by RE practitioners.

Purpose:
The dataset supports transparency, reproducibility, and further analysis of AI-based methods for software requirements quality assessment and can serve as a foundation for future benchmark datasets and standardized evaluation frameworks.