A Comparative Study on Image Recommendation Using Siamese Neural Network Architecture

In recent years, many e-commerce platforms like Amazon, Myntra, Adidas, Jumia, Temu, and others have adopted advanced recommendation systems to suggest clothing based on customer preferences. The goal is to quickly satisfy consumers while maximizing company revenues. These technologies aim to enhance the shopping experience and increase customer loyalty. Personalized recommendations help optimize sales and attract more buyers. In this work, we propose an innovative image recommendation method based on Siamese networks. The aim of this approach is to analyze the similarity between different clothing items and group similar images in the same latent space. This enhances the matching of images with common characteristics while optimizing the response time of the recommendation system. We designed a recommendation model based on Siamese networks and CNN architecture, comparing it with the most commonly used transfer learning models in the field of image recommendation. The aim of this study is to compare the performance of CNN architecture with that of transfer learning techniques. The results show better precision with a score of 0.9669 accuracy for ResNet50V2, 0.9625 for CNN, and 0.9571 for VGG19, achieved on the FashionMNIST dataset.