Automated requirements assessment traditionally relies on universal patterns as proxies for defectiveness, implemented through rule-based heuristics or machine learning classifier strained on large annotated datasets. However, what constitutes a 'defect' is inherently context-dependent and varies across projects, domains, and stakeholder interpretations. We propose a Human-LLM Collaboration approach that treats defect prediction as an adaptive process rather than a static classification task. To probe the feasibility of this approach, we conducted an empirical investigation of different LLM-based classification strategies on an industrial requirements dataset. We derived practical design implications for tool-support and developed our first prototype: Requirely lets users flexibly configure checkers, shows explanations for its predictions, and provides automatically generated improvement suggestions to fix findings inanintegrated environment.
Max Unterbusch (Thu,) studied this question.