信息熵方法评估海关生成式人工智能研究  

Research of the Evaluation of Customs Generative Artificial Intelligence with Information Entropy

作　　者：王翔[1,2] WANG Xiang(The National Information Center,General Administration of Customs,Beijing 100005;Laboratory of International Trade IT Standards,General Administration of Customs,Beijing 100005)

机构地区：[1]全国海关信息中心,北京100005 [2]海关国际贸易信息标准化应用创新实验室,北京100005

出　　处：《中国口岸科学技术》2024年第6期4-9,共6页China Port Science and Technology

基　　金：海关总署科研项目(2019HK018、2020HK281、2020HK300、2022HK053)。

摘　　要：世界海关组织(World Customs Organization,WCO)报告显示,生成式人工智能(Generative Artificial Intelligence,GAI)将逐步成为各国海关的技术选择。以信息熵作为量化评估措施,可以分析不同GAI场景的适应性,比较海关不同合作方案中的技术架构。分析显示,单独采用外部经贸语料库或通用语料库,在考虑资源投入和数据安全性的前提下,适用的应用场景较少,仅大致符合“调查研究”1个场景的需要;通过组合内外部合作方式,查询外部GAI模型获得特征标签、参数的方式,不仅资源投入较低,也能够大致覆盖WCO列举的各个场景。借助必要的微调措施,可以将GAI融入海关信息化整体环境,构建GAI为基础的新型生产力。According to a report bythe WCO,the Customs community will increasingly adopt generative artificial intelligence(GAI)technology.Information entropy,as a quantitative evaluation metric,can be used to assess the adaptability of various GAI scenarios and compare alternative technologicalframeworks under different Customs collaboration models.The results demonstrate that employing external economic and trade corpus or general corpus alone,while taking into consideration resource investment and data security,results in limited applicable scenarios,roughly meeting the needs of only one scenario labeled as"investigation and research".However,with the collaboration of internal and external artificial intelligence(AI)resources,particularly by querying external GAI models to obatain domain tags and parameters,not only is the resource investment relatively low,but it can also roughly cover the many scenariosenumerated by the WCO report.With proper fine-tuning,GAI can be integrated into Customs'IT environment,allowing for the creation of new types of productive forces based on GAI.

关 键 词：智慧海关 信息熵 大模型(LM) 生成式人工智能(GAI) 世界海关组织(WCO) 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] F752.5[自动化与计算机技术—控制科学与工程] F49[经济管理—国际贸易]