Agent4RE：面向端到端软件需求工程的自精炼多智能体框架与基准
原标题：Agent4RE: A Self-Refining Multi-agent Framework for End-to-End Software Requirements Engineering and Benchmarking
AI 摘要
研究团队提出 Agent4RE，一个用于端到端软件需求工程（RE）的自精炼多智能体框架，通过编排专门智能体并引入两个迭代改进循环，覆盖从需求获取到生成的全流程。为支持评估，团队构建了基于真实人工需求规格的 RE-E2E 端到端基准数据集。在 8 个 LLM 上的评测显示，三种 Agent4RE 变体在文本指标上平均超越领域上下文增强提示基线 8%，两个增强变体在 LLM-as-a-judge 和人工评分上比两个 RE 基线高出约 0.8 分（四分制）。
正文节选
Agent4RE: A Self-Refining Multi-agent Framework for End-to-End Software Requirements Engineering and Benchmarking Abstract. Existing LLM-based approaches for software Requirements Engineering (RE) typically rely on basic prompting strategies or rudimentary agent collaboration, under-utilizing the full potential of multi-agent systems. Meanwhile, available datasets focus on isolated subtasks, such as requirements extraction, classification, and completeness detection, leaving the absence of an en