EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

867. Classifying import containers based on cargo contents: an unsupervised text classification

Invited abstract in session MA-3: Industrial Optimization, stream Data Science Meets Optimization.

Monday, 8:30-10:00
Room: 1005 (building: 202)

Authors (first author is the speaker)

1. Ying Xie
School of Management, Cranfield University
2. Dongping Song
School of Management, University of Liverpool

Abstract

Post unloading from vessels, import containers are temporarily stacked at seaport yards before being transported to inland customers. A significant challenge faced at container ports is the optimal storage of import containers, including proximity to out-terminals and minimizing reshuffling needs. However, uncertainties surrounding collection times and transportation mode pose challenges for port operators. Nevertheless, port operators possess knowledge of each import container’s contents upon unloading from vessels. An interesting research question arises:can cargo content information be leveraged to enhance container storage management? Addressing this question entails categorising import containers into manageable classes based on cargo contents. This study tackles the challenge of categorising a high variety of cargo contents, often exceeding 100K unique items, using the Standard International Trade Classification as labels. Our novel unsupervised text classification approach employs pretrained Glove Word Embeddings and Cosine Similarity for label assignment. This integration offers an efficient approach to generate training datasets in data-scarce scenarios. Additionally, we address data imbalance through three strategies:upsampling/downsampling, class weights, and weighted loss function. Leveraging the Transformers based neural-network models on enriched and balanced datasets, we achieve promising results in accurately classifying container content.

Keywords

Status: accepted

Back to the list of papers