Automating Harmonized System (HS) Code Classification from Unstructured Shipping Manifests using Large Language Models

This study proposes a structured framework for automating Harmonized System (HS) code classification from short-text product descriptions, using natural language processing (NLP). The technical architecture is designed to address key challenges in processing unstructured shipping manifest data, including linguistic ambiguity and inconsistent formatting. Multiple machine learning approaches and Large Language Models (LLMs) are evaluated, with the top performing model achieving 87.9% classification accuracy at the 6-digit HS code level. The research aims to streamline customs classification procedures with a scalable solution. Beyond streamlining customs procedures, the framework enables ancillary applications such as trade analytics and enhanced compliance monitoring.