Reproduce “AesRec: A Dataset for Aesthetics-Aligned Clothing Outfit Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenxin Ye, Lin Li, Ming Li, Yang Shen, Kanghong Wang, Jimmy Xiangji Huang
A new large-scale dataset, AesRec, integrates over one million user-outfit interactions with systematic, multidimensional aesthetic scores generated by Vision-Language Models (VLMs) for clothing items and outfits. It enables recommendation systems to provide explicit aesthetic guidance, demonstrating a significant improvement in the aesthetic quality of recommendations and a favorable shift towards higher-aesthetic content in ranking exposure compared to traditional preference-based models.
Key insights
- Advanced Vision-Language Models (VLMs) can reliably provide large-scale, multidimensional aesthetic annotations for fashion, offering a scalable and cost-effective alternative to traditional manual labeling.
- Explicitly incorporating quantified aesthetic information into recommendation algorithms significantly enhances the aesthetic quality of recommendations and favorably biases ranking exposure towards aesthetically superior items.
- Traditional interaction-based recommendation models, without explicit aesthetic guidance, often fail to improve aesthetic quality and can even exhibit a systematic bias towards lower-aesthetic content in their recommendations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.