Reproduce “Query-dominant User Interest Network for Large-Scale Search Ranking” 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
Tong Guo, Xuanping Li, Haitao Yang, Xiao Liang, Yong Yuan, Jingyou Hou, Bingqing Ke, Chao Zhang and 4 more
Researchers from Kuaishou Technology developed the Query-dominant User Interest Network (QIN) to enhance personalized search ranking by effectively leveraging users' comprehensive interaction histories. QIN outperforms existing methods on industrial datasets, achieving a 7.6% improvement in click-through rate in production.
Key insights
- A two-stage relevance search process effectively filters users' extensive behavior history, first by query relevance then by target item relevance, to address noise.
- An adaptive fusion mechanism in the Fused Attention Unit dynamically balances the importance of ID and attribute embeddings based on user engagement levels.
- Performance generally improves with longer behavior sequences up to 10-15 items, and a balanced consideration of ID and attribute information (alpha around 0.4-0.5) optimizes results.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.