Product Recommendation in Fashion E-Commerce using Deep Learning and Image Similarity

This is the authors' abstract. We don't add key points for this paper.

This research proposes a fashion product recommendation system using contentbased image retrieval (CBIR) with convolutional neural networks and the k-nearest neighbors (KNN) algorithm.Product images from an e-commerce website were collected through web scraping.Pre-trained VGG16 extracted image features.KNN identified visually similar products based on feature vectors.Cosine similarity, root mean square deviation (RMSE), and structural similarity index (SSIM) evaluated model performance.On a test set, the model achieved an average cosine similarity of 0.76 and SSIM of 0.80 between query and recommended items.The system was deployed on a website, demonstrating the feasibility of content-based visual recommendations for fashion e-commerce.This research makes a valuable contribution by developing and evaluating a domain-specific recommender system to enhance the online shopping experience.