AI 소프트웨어 개발을 위한 양방향 수렴 기반 중간 산출물 복원을 통한 요구사항 추적성

초록

With the recent shift toward development methods utilizing generative AI and AI agents, the software industry is experiencing accelerated development. However, this approach does not generate any intermediate artifacts—such as use cases, use case scenarios, and object artifacts—between requirements and source code, resulting in a critical problem where requirement traceability cannot be established. To address this issue, we propose a technique that utilizes Large Language Models (LLMs) to recover these missing intermediate artifacts and assign trace links during the recovery process. The proposed method performs forward recovery, which generates use cases solely from requirements, and backward recovery, which recovers use cases solely from source code. This bidirectional convergence approach increased use-case recovery coverage from 65.5% to 75.1%, a gain of 9.6 percentage points, and improved trace-link recall.

키워드