<![CDATA[Requirements elicitation is a critical activity in software engineering, as incomplete, ambiguous, or inconsistent requirements can adversely affect software quality and project outcomes. Although Large Language Models (LLMs) have demonstrated considerable potential for supporting Requirements Engineering, conventional prompting approaches typically rely on direct or predefined interactions and lack systematic mechanisms for identifying and resolving information gaps during elicitation. This study proposes Adaptive Context-Driven Prompting (ACDP), an iterative framework designed to improve LLM-supported requirements elicitation through context preservation and gap-driven refinement. ACDP progressively constructs an Elicitation Context (EC) that maintains project objectives, stakeholders, constraints, assumptions, and previously elicited requirements. In parallel, a Requirement Gap Matrix (RGM) systematically identifies missing, partial, ambiguous, and conflicting information and transforms these gaps into targeted re-elicitation prompts for subsequent LLM interactions. The framework was evaluated using public software requirements documents from the PURE dataset and compared with Zero-Shot Prompting and Fixed Multi-Step Prompting under controlled LLM settings. Performance was assessed using reference-based Precision, Recall, and F1-score, together with requirement-quality dimensions including completeness, correctness, consistency, clarity, relevance, ambiguity, and redundancy. Experimental results demonstrate that ACDP achieved a Precision of 0.93, Recall of 0.92, and F1-score of 0.92, outperforming both baseline prompting approaches while also improving requirement completeness, correctness, and overall quality. These findings demonstrate that combining persistent elicitation context with explicit gap diagnosis provides a systematic and effective mechanism for improving LLM-assisted requirements elicitation and offers practical support for requirements analysts in complex software projects.]]>
