Women’s Outfit Recommendation System for Fashion Product Personalization Using the BERT4Rec Method

Abstract

The fashion industry today focuses on understanding individual customer preferences rather than merely following global trends. The massive growth of e-commerce platforms and the emergence of fast fashion have exposed consumers to a vast range of outfit choices, making it challenging to identify products that match their personal style. This study aims to design and implement a women’s outfit recommendation system using the BERT4Rec (Bidirectional Encoder Representations from Transformers for Sequential Recommendation) method to provide relevant and personalized product recommendations based on user interaction sequences. The research uses the ModCloth public dataset containing user–item interactions and women’s outfit attributes. The process includes data preprocessing, user sequence generation, model training using the Transformer encoder architecture, and performance evaluation using Hit Ratio (HR@K) and Normalized Discounted Cumulative Gain (nDCG@K). Experimental results on the ModCloth dataset show that the BERT4Rec model effectively captures user sequential behavior and achieves strong recommendation accuracy, with HR@10 = 0.2532 and nDCG@10 = 0.1293. These results indicate the model’s capability to predict user preferences with significant accuracy and to generate contextually relevant outfit suggestions. The findings highlight the potential of BERT4Rec in supporting personalization across e-commerce platforms, virtual fashion assistants, and intelligent retail systems, paving the way for adaptive fashion recommendation technologies.