Reproduce “PIER: Permutation-Level Interest-Based End-to-End Re-ranking Framework 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
Xiaowen Shi, Fan Yang, Ze Wang, Xiaoxu Wu, Muzhi Guan, Guogang Liao, Yongkang Wang, Xingxing Wang and 1 more
The PIER framework addresses re-ranking limitations in e-commerce by introducing a two-stage architecture that models user interests at a permutation level. This approach improved Click-Through Rate and Gross Merchandise Volume in real-world deployment on Meituan's food delivery platform.
Key insights
- Considering user interest at the permutation level, rather than just item-level interest, allows for more effective personalization in re-ranking.
- Efficiently selecting promising permutations from a large search space can be achieved by leveraging SimHash techniques to match historical user behaviors with candidate permutations.
- A fine-grained, omnidirectional attention mechanism can effectively model both inter-item and intra-feature relationships within a permutation to predict user interaction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.