Reproduce “A Unified Search and Recommendation Framework Based on Multi-Scenario Learning for Ranking in E-commerce” 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
Jinhan Liu, Junjie Xu, Junjie Li, Baoli Li, Sulong Xu
JD.com researchers developed the Unified Search and Recommendation (USR) framework to enhance multi-scenario learning for e-commerce ranking. It integrates fine-grained, scenario-specific user interests and features with a global view of user actions, demonstrating significant gains in user conversion and click-through rates when deployed on the 7Fresh App.
Key insights
- Explicitly distinguishing and modeling scenario-specific user interests and adaptively scaling scenario-agnostic features are crucial for effective joint optimization of search and recommendation systems.
- Incorporating a 'global label space' by modeling the probability of an action given a global action offers a more holistic and powerful supervisory signal for multi-task learning across distinct scenarios.
- Contrastive learning successfully separates and clusters representations for different scenarios, validating its utility in learning distinct features and interests for search and recommendation tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.