INTEGRATING KNOWLEDGE GRAPHS AND LLM FOR INDONESIA HS CODE CLASSIFICATION: A FOCUS ON HEAVY EQUIPMENT PARTS

Abstract

Accurate classification of heavy equipment parts under Indonesia’s Harmonized System (HS) codes is vital for regulatory compliance, cost mitigation, and operational efficiency. This paper presents a proof-of-concept framework that integrates Knowledge Graphs (KGs), Large Language Models (LLMs), and Machine Learning (ML) techniques to address the persistent challenges of HS code misclassification. The framework leverages historical customs data to train ML models for HS code prediction and incorporates a Knowledge Graph to encode domain-specific rules and hierarchical relationships. Complementing these methods, LLMs provide contextual understanding and generate explainable justifications for classification decisions. As a proof-of-concept, the system is demonstrated through a case study at a leading enterprise in Indonesia’s heavy equipment sector which exhibits enhanced accuracy, transparency, and operational efficiency in HS code assignment. Accounting for nuanced part descriptions, evolving regulations, and the inherent complexity of hierarchical classification, the proposed approach minimizes reliance on implicit knowledge and reduces error rates. The findings highlight the framework’s scalability and adaptability, indicating its potential to benefit not only the heavy equipment industry but also other sectors that confront similar global trade compliance challenges.