Application of machine learning for assessment of HS code correctness

Abstract

Manual assessment of the correctness of Harmonized System codes of goods is very error-prone and time demanding task taking into account the dramatically growing amounts of cross-border trade. The paper provides an automated solution to this problem by applying machine learning methods to assess the correctness of Harmonized System codes. We use machine learning for providing predictions and recommendations of Harmonized System codes on the basis of a model learned from the textual descriptions of the products. In order to assess the correctness Harmonized System codes of goods we introduce a novel combined similarity measure based on cosine similarity of texts and semantic similarity of Harmonized System codes calculated according to their taxonomy. We also present and prove the properties of this new similarity measure. We test our method on the real open source data set of Bill of Lading Summary 2017.