IRAP: Interactive Retrieval-Augmented Preference Elicitation for Quantifying Software Performance Requirements
Researchers have formalized the problem of quantifying software performance requirements from natural language into mathematical functions, proposing IRAP (Interactive Retrieval-Augmented Preference Elicitation). The approach explicitly derives problem-specific knowledge to retrieve and reason preferences, guiding progressive stakeholder interaction while reducing cognitive overhead. Experiments against 10 state-of-the-art methods on four real-world datasets demonstrate IRAP's superiority across all cases. The work addresses the challenge of ambiguity in performance requirements and uncertainty in human cognition, which have hindered automated quantification. The paper is published on arXiv with ID 2604.21380.
Key facts
- IRAP quantifies performance requirements into mathematical functions via interactive retrieval-augmented preference elicitation.
- The approach explicitly derives problem-specific knowledge to retrieve and reason preferences.
- It guides progressive interaction with stakeholders while reducing cognitive overhead.
- Experiments were conducted against 10 state-of-the-art methods on four real-world datasets.
- IRAP demonstrated superiority on all cases.
- The paper formalizes the problem of quantifying software performance requirements from natural language.
- The work addresses ambiguity in performance requirements and uncertainty in human cognition.
- The paper is available on arXiv with ID 2604.21380.
Entities
Institutions
- arXiv