Machine Learning for Solving Items Classification Problems in E-com Services

In the rapidly growing e-commerce industry, as in any business, competitor analysis plays an important role, the results of which can influence business decisions. Assortment analytics in conjunction with competitor data provides strong support to the commercial unit by providing recommendations for category management. This work is aimed at creating an algorithm that allows to determine the category of a product by only its name, which allows to denote competitors' products according to a specific category tree, using machine learning classification methods. The study uses real data from one of the largest e-grocery services. Data preprocessing techniques, embedding, and CatBoost were applied to create, train, and evaluate the models. The model was successfully integrated into the company's relevant processes, saving 33.5 man-hours in manually marking up every 10,000 items. In addition, the work also explored methods related to neural networks, such as CNN for example, which increased the category prediction accuracy to 0.97 and identified them as the most effective way to implement the solution.