Development of a system for creating and recommending combination collections in the e-commerce clothing industry

Abstract

In the clothing sector, matching the right demand with the appropriate user is of great significance. Combination suggestions emerge as an innovative strategy for e-commerce platforms operating in the clothing sector. By providing suitable combination suggestions tailored to the right user, the profit margin of sales increased, and the brand image strengthened. The aim of this study is to develop a recommendation system based on image processing and machine learning that generates combinations from products that may interest users and recommends these combinations to them. 90 million possible combinations have been obtained using a dataset consisting of products detected from images of items sold in the clothing category on Trendyol. These combinations have been trained using the Prod2Vec algorithm to create new pairings. Subsequently, collections have been developed for purchasing looks using image processing methods. In this context, the You Only Look Once (YOLO) model has been selected for clothing classification, while the Convolutional Network Next (ConvNext) model has been employed for calculating image similarity. Models have also been developed for estimating click performance using Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Linear Regression (LR). The prediction performances of the developed models have been evaluated using Coefficient of Determination (R2), Mean Squared Error (MSE), and Mean Absolute Error (MAE) metrics. When the developed models have been examined, it has been observed that RF had superior performance. The developed system provided a 5% increase in the time spent on the Trendyol mobile application.