Research And Implementation Of Cross-border HSCode Prediction System Based On Knowledge Graph

In recent years,e-commerce has gradually become an important focus to promote the domestic cycle and smooth the domestic and international dual cycle.With the development of e-commerce economy,cross-border commodity transactions have ushered in a period of vigorous development,making China the world’s largest exporter.When carrying out import and export trade,merchants are required to provide the HS code of import and export commodities to the customs.HS code is an international general classification system for import and export commodities formulated by the International Customs Council.It is a system for quantitative management of various entry and exit tariff rates of different commodities.At present,more than 98%of global trade volume follows this system.HS code has become a standard language in international trade.The complex structure of HS code easily leads to errors,so its classification,query and confirmation require a lot of manpower and professional business knowledge.At present,HS code has become one of the important factors affecting the efficiency of international trade goods clearance.How to get HS code automatically through the text description of goods has become an urgent problem to be solved.In this study,in order to solve a series of problems of HS code,including cumbersome retrieval process and low efficiency,the research and design of cross-border commodity HS code prediction system based on knowledge graph was carried out.The main work includes:1.In view of the difficulty in obtaining data in the field of HS code and the loss of structural information caused by the use of traditional commodity text representation,the data was crawled and analyzed from four public websites and one official document,the HS code data set containing more than 400000 pieces of data was constructed,a novel knowledge graph method was designed,and the cross-border commodity HS code knowledge Graph based on inclusion relationship was constructed.2.In view of the problem of low quality and efficiency of HS code retrieval,a heterogeneous graph neural network HS code prediction model based on the Subgraph classification method on the knowledge graph was proposed.Through the experimental comparison with other methods,its effectiveness and efficiency were verified.3.In view of the problem of insufficient query information input by users in reality,a HS code prediction model based on active interaction was proposed,which can interact with users to supplement commodity information.In addition,the fact that the model can improve the prediction accuracy was proved by experiments.In this study,based on the cross-border commodity HS code knowledge graph,heterogeneous graph neural network HS code prediction model and HS code prediction model with active interaction function were integrated,and the cross-border commodity HS code prediction system was developed.System includes three modules,namely,information retrieval,results display,active interaction,which has not only realized the automatic retrieval function of HS code based on commodity text description and the active interaction function in case of insufficient information,but also met needs of cross-border transaction merchants for HS code retrieval,and greatly improved the work efficiency of cross-border transaction merchants.