Categorising Products in an Online Marketplace: An Ensemble Approach
This addresses the problem of automated product categorization for e-commerce companies, but it is incremental as it builds on existing ensemble methods.
The study tackled product categorization in online marketplaces by proposing an ensemble approach combining XGBoost and k-nearest neighbors to predict categories, subcategories, and colors, achieving an average F1-score of 0.82.
In recent years, product categorisation has been a common issue for E-commerce companies who have utilised machine learning to categorise their products automatically. In this study, we propose an ensemble approach, using a combination of different models to separately predict each product's category, subcategory, and colour before ultimately combining the resultant predictions for each product. With the aforementioned approach, we show that an average F1-score of 0.82 can be achieved using a combination of XGBoost and k-nearest neighbours to predict said features.