Reproduce “Universal Multi-modal Multi-domain Pre-trained 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
Wenqi Sun, Ruobing Xie, Shuqing Bian, Wayne Xin Zhao, Jie Zhou
UniM2Rec is a framework for recommendation systems that integrates multi-modal item information (text, images) and leverages cross-domain user behaviors to generate universal item representations. It consistently outperforms state-of-the-art baselines in Recall and NDCG across five datasets, demonstrating robustness to missing modality data and effectiveness in handling few-shot items.
Key insights
- Leveraging both multi-modal item content and cross-domain user behaviors is crucial for building comprehensive and transferable user and item representations.
- A universal pre-training approach allows models to learn robust representations that bridge semantic gaps and transfer knowledge effectively across diverse domains and modalities.
- Employing a modular architecture with frozen pre-trained encoders and adaptable projectors enhances performance while maintaining robustness to real-world data imperfections like missing modalities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.