Automated Prioritization of Non-Functional Requirements Using Large Language Models: A Source Code-Based Approach

Research Context: Non-Functional Requirements (NFRs) are critical for ensuring the quality of Information Systems, especially in domains where failures in performance or security may cause severe consequences. In organizational contexts, however, eliciting and prioritizing NFRs is challenging due to scarce documentation and reliance on subjective judgments. This work addresses the intersection of processes, technologies, and organizational needs by automating NFR analysis.

Practical Problem: Traditional prioritization methods, such as AHP or MoSCoW, depend heavily on stakeholder input and rarely address NFRs systematically, limiting applicability to poorly documented or large-scale systems and exposing organizations to quality risks. 

Proposed Solution: We propose a source code--based pipeline that automatically extracts, classifies, and prioritizes NFRs using Large Language Models (LLMs). The approach employs semantic analysis of code snippets guided by structured prompts, producing standardized NFR descriptions, categories aligned with ISO/IEC 25010, and priority levels (High, Medium, Low). 

Related IS Theory: The study draws on theories of Software Quality and Requirements Engineering, highlighting quality models (ISO/IEC 25010), decision-making approaches for prioritization, and their organizational impact. 

Research Method: An empirical evaluation was conducted on the OpenMRS repository (134 Java files). The automated pipeline used GPT-based semantic analysis, validated against manual annotations from three experts. Performance was assessed with precision, recall, and F1-score. 

Summary of Results: The pipeline extracted 422 NFRs versus 141 manually identified, achieving 93.4% recall, 30.3% precision, and a 45.8% F1-score. High-priority NFRs were mainly associated with Security and Performance Efficiency, reflecting the system's critical nature.