# "Does it come in black?" CLIP-like models are zero-shot recommenders

**Authors:** Patrick John Chia, Jacopo Tagliabue, Federico Bianchi, Ciro Greco, Diogo Goncalves  
**arXiv:** [2204.02473](https://arxiv.org/abs/2204.02473)  
**Date:** 2022-04-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2204.02473v2/

## Key points

- CLIP-like models are zero-shot recommenders
- Current item-to-item recommendation systems do not allow users to explore changes along selected dimensions
- CLIP-based models can suggest items similar to a query item but with different attributes such as color
- The authors propose using CLIP-based models as zero-shot recommenders for comparative item recommendations
- GradREC, a large model built for fashion, effectively supports this use case
- GradREC has the potential to enhance product discovery and improve the online shopping experience
- The paper examines the industry potential of GradREC and assesses its strengths and weaknesses
- Comparative recommendations are significant in online shopping platforms
- CLIP-based models like GradREC fulfill the need for comparative recommendations efficiently

## Summary

CLIP-like models are zero-shot recommenders, and authors Patrick John Chia, Jacopo Tagliabue, Federico Bianchi, Ciro Greco and Diogo Goncalves discuss the importance of product discovery in online shopping. They emphasize that current item-to-item recommendation systems do not enable users to explore changes along selected dimensions. For instance, if a user wants to find something similar to a query item but in a different color, existing models cannot provide suitable recommendations. To address this limitation, the authors propose using CLIP-based models as zero-shot recommenders for item recommendations of a comparative nature. These models can suggest items that are similar to the query item but differ in specific attributes such as color. The authors demonstrate how leveraging a large model built for fashion called GradREC can support this use case effectively. The paper also examines the industry potential of GradREC and provides an initial assessment of its strengths and weaknesses. By allowing users to explore changes along selected dimensions, GradREC has the potential to enhance product discovery and improve the overall online shopping experience. Overall, this paper highlights the significance of incorporating comparative recommendations into online shopping platforms and showcases how CLIP-based models like GradREC can fulfill this need efficiently. The findings presented contribute to advancing recommendation systems in e-commerce and offer valuable insights for researchers and industry professionals alike.