Researchers have developed UniRec, a novel model designed to improve cascaded recommender systems by fusing information across different stages of the recommendation process. Unlike previous methods that focused on single-stage fusion or simple cross-stage coordination, UniRec employs a unified computation graph with shared embeddings and a dual-axis preference alignment objective. This approach ensures consistency between upstream and downstream stages and reorganizes pairwise objectives into bidirectional preference evidence. The model also incorporates attribute group-relative regularization to prevent over-concentration on high-reward regions. Deployed on the Kuaishou platform, UniRec has demonstrated offline improvements over baseline methods and achieved a 0.616% gain in app usage duration during online A/B tests. AI
IMPACT Enhances recommender system performance by improving cross-stage fusion and preference alignment, potentially leading to more engaging user experiences.
RANK_REASON Publication of a research paper on a new model for recommender systems. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
