Building framework recommendation system for trendy fashion e-commerce based on deep learning with Top-K

Ho Thanh Thuy,Pham The Bao,Do Dieu Le
DOI: https://doi.org/10.30574/ijsra.2024.12.2.1270
2024-07-30
International Journal of Science and Research Archive
Abstract:Recently, e-commerce has become a vital component of our purchasing habits. Central to this evolution is the recommendation system, an advanced algorithm designed to personalize the shopping experience and significantly boost consumer demand. With its diverse and ever-changing inventory, the fashion industry benefits immensely from these algorithms, making it a fascinating case study for understanding the broader impacts of technology on consumerism. Traditional fashion recommendation systems are fundamentally based on item compatibility, but keeping up with trends is also essential. To address this, we propose a two-stage system: fashion detection and outfit suggestions based on the identified items. Users receive images of Key Opinion Leaders (KOLs) or Influencers wearing similar outfits. These recommendations ensure item compatibility, offer diverse styles, and remain fashionable. At the outset, we experimented with YOLOv8 to select the best version. Next, we implemented fashion image retrieval based on feature extraction using two pre-trained networks. To enhance reliability, we developed a voting and ranking algorithm. Our experiments, conducted on a self-collected dataset, evaluated the system’s effectiveness in detecting fashion objects and the efficiency of content-based image retrieval
What problem does this paper attempt to address?