Development of a Fashion Outfit Recommendation System Using Deep Feature Extraction and Graph-Based Learning

In online fashion retail, the demand for automated recommendation systems that help users create compatible clothing outfits is steadily increasing. This study presents a graph-based outfit recommendation system to reduce workload, optimize inventory management, and enhance user experience. Product images are segmented using the SegFormer model to separate each fashion item, and multi-layered features are extracted using the ResNet50 architecture. Product similarities are calculated using the cosine similarity metric to determine the most suitable outfits. The proposed system models fashion products as nodes and their relationships as edges, forming a graph-based structure. This approach automates the outfit creation process, reducing workload and saving time. The results demonstrate that the system successfully generates compatible outfits based on both style and category and has the potential to contribute to inventory management.