Current Location

You are here: World Customs Organization   Media   Newsroom   2022   March   WCO BACUDA experts develop a neural network model to assist classification of goods in HS

WCO BACUDA experts develop a neural network model to assist classification of goods in HS

03 March 2022

In line with the WCO's recognition of data analytics and machine learning as an integral part of the future of customs, The WCO BACUDA project was launched in September 2019 as a collaborative research platform focused on data analytics. At the beginning of 2020, BACUDA's expert group collaborated on the development of an HS Code Recommendation AI, which aims to solve difficulties arising during the classification task for both traders and customs officials by introducing a model that uses historical data to recommend HS codes for commercial descriptions of goods.

The collaboration with Nigeria Customs Service, which voluntarily provided import data for the project, commenced with joint work on HS code recommendation modelling. This choice was premised on the observation that difficulty in interpreting complex commodity nomenclature descriptions is one of the factors that make commodity classification a burdensome task and causes unintentional misclassification.

The model employs state-of-the-art natural language processing technology based on an artificial neural network called Doc2Vec. This technology enables the model to recognize the semantic relationship between words in commodity descriptions and their relationship to HS codes and recommend accurate HS codes for new commodity descriptions or those not present in the nomenclature. In addition, the model is highly optimized for trader-declared description data, as it incorporates various techniques for pre-processing the data.

How does the model work?

The basic idea of the model is to learn the pairs of HS codes and commercial descriptions of goods declared by traders. Based on the learning, the model recommends several HS codes when a new description is entered. During the learning process, HS codes and commodity descriptions are converted into numerical values in the vector space while preserving their semantic relationships. When a new description is entered, the model calculates its numerical value and recommends HS codes that have the most similar values.

Technical support

For further customized support, WCO invites Members to contact WCO BACUDA project team (bacuda@wcoomd.org) to organize a joint test of the HS Recommendation AI model with BACUDA experts.