Меню
Vol. 4 No. 29 (2025) / 2025-12-30 / Number of views: 153
Authors
Keywords
Link to DOI:
How to quote
With the growth of e-commerce, marketplaces have become key platforms for interaction between buyers and sellers. Every day, a large number of new products are added, requiring categorization to enable efficient assortment management, improve search accuracy, and generate personalized recommendations. This study presents a product categorization module developed in Python 3.10.10, which addresses the task of classifying products based on their textual descriptions (104 026 products, 62 categories). A comprehensive analysis was conducted on classical algorithms (Random Forest, SVM, Logistic Regression, Multinomial Naïve Bayes) and neural network architectures (RNN, CNN, Transformer), as well as an ensemble approach (Random Forest + Decision Tree + Logistic Regression). Data preprocessing included tokenization, removal of stopwords and non-alphabetic characters, stemming, and text normalization. Experiments on a test set (20 805 entries) showed that Random Forest achieved an accuracy of 94.2% with a processing time of 74.9 ms, outperforming neural network models (RNN - 91.6%, Transformer - 90.9%) and the ensemble method (94.3% at 75.2 ms). The key scientific contribution of this work lies in demonstrating the applicability of classical methods for medium-complexity tasks under resource constraints. The results show that Random Forest is an optimal solution for integration into real-world marketplace systems, providing high accuracy without significant computational overhead.