摘要
协调制度（HS）关税分类是一项高风险专家任务，难点在于需同时满足材料、形式、功能等多轴线的竞争性优先规则。针对大模型端到端提示在多维规则推理上的失效，本文提出一种确定性代理工作流。该架构采用固定控制流，将模型调用限制在狭窄阶段，并保留局部反思与验证机制，从而通过结构化输出和原文引用实现决策的可解释性。实验表明，该方法在六位数分类任务中显著优于基线，且开源模型表现接近前沿模型。
AI 推荐理由
论文核心解决多维规则推理难题，通过确定性工作流实现可解释的逻辑推导。
研究机构
School of Information and Electronic Engineering, Shanghai Jiao Tong University, Shanghai, China
Customs National Supervision Bureau for Duty Collection (Shanghai), General Administration of Customs of the P.R.C. (GACC), Shanghai, China
Nanjing Jiyun Information Technology Co., Ltd., Nanjing, China
School of Computer Science, Shanghai Jiao Tong University, Shanghai, China
Department of Science and Technology (Shanghai), General Administration of Customs of the P.R.C. (GACC), Shanghai, China
论文信息