Large Language Models | TransportXhttps://transportxlab.com/tag/large-language-models/Large Language ModelsHugo Blox Builder (https://hugoblox.com)en-usWed, 01 Jul 2026 00:00:00 +0800https://transportxlab.com/media/icon_hu11734318148517933569.pngLarge Language Modelshttps://transportxlab.com/tag/large-language-models/HSGraphAgent: Knowledge-Graph-Guided Large Language Models for Harmonized System Code Classificationhttps://transportxlab.com/publication/hsgraphagent/Wed, 01 Jul 2026 00:00:00 +0800https://transportxlab.com/publication/hsgraphagent/<h2 id="研究概览">研究概览</h2> <p>HSGraphAgent 将海关 HS 编码体系建模为包含层级关系和排除规则的知识图谱，引导大语言模型从 Section、Chapter、Heading 到 Subheading 逐级完成商品归类。该方法强调规则约束与可解释推理，避免仅凭文本相似度选择编码。</p> <h2 id="核心方法与发现">核心方法与发现</h2> <ul> <li>Select 阶段只在当前节点的合法子节点中选择，保证层级路径有效。</li> <li>Redirect 阶段在触发税则排除条件时动态修正推理路径。</li> <li>在 6 位 HS Code 细粒度任务上，图谱引导的分层推理明显优于直接生成和常规 RAG 基线。</li> </ul> <h2 id="资料来源">资料来源</h2> <ul> <li><a href="https://doi.org/10.18653/v1/2026.acl-long.2072" target="_blank" rel="noopener">ACL Anthology DOI</a></li> <li><a href="https://github.com/VoldeMordddddd/HSBench" target="_blank" rel="noopener">HSBench 代码与数据</a></li> <li><a href="https://mp.weixin.qq.com/s/RV14fJCbhqTD2RtKYkaPwQ" target="_blank" rel="noopener">TransportX Lab 公众号解读</a></li> </ul>