Less is More: A Streamlined Graph-Based Fashion Outfit Recommendation without Multimodal Dependency

Our work introduces a novel approach, Tripartite Graph Integration for Fashion outfit recommendation (TGIF). Unlike previous studies relying on image embeddings, TGIF leverages only graph structure information, reducing resource-intensive processes. The model constructs an integrated graph by adding the user-item relationship to the user-outfit-item tripartite graph. Utilizing subgraph-based graph neural networks and target node-aware attention pooling, TGIF achieves high compatibility and recommendation performance without using images or text. Experimental results on real-world data demonstrate superior performance over state-of-the-art models in personalized outfit recommendation and out-fit compatibility matching tasks. The source code is available at https://github.com/venzino-han/TGIF_SAC2024.