Reproduce “MBGR: Multi-Business Prediction for Generative Recommendation at Meituan” 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
Changhao Li, Junwei Yin, Zhilin Zeng, Senjie Kou, Shuli Wang, Wenshuai Chen, Yinhua Zhu, Haitao Wang and 1 more
Meituan's MBGR framework, a generative recommendation system tailored for multi-business platforms, integrates business-aware representations and prediction modules to mitigate issues like "representation confusion" and the "seesaw phenomenon." This framework demonstrated a +3.98% improvement in weighted average CTCVR during online A/B testing on Meituan's food delivery platform, showing particular gains for smaller business categories.
Key insights
- Explicitly incorporating business context into semantic ID learning and item prediction through BID and MBP modules effectively mitigates "representation confusion" and the "seesaw phenomenon" in multi-business settings.
- Dynamically routing and weighting labels via LDR, combined with optimized business-specific weights, significantly enhances multi-business generation capability, especially for smaller business categories.
- A multi-objective loss function, combining contrastive learning with semantic reconstruction, enables robust end-to-end training that balances prediction accuracy with semantic preservation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.