Researchers have developed OGR, a novel end-to-end framework for slate recommendation that generates and ranks item lists simultaneously. The system, detailed in an arXiv paper, introduces TUSID to create hierarchical semantic IDs by fusing item-specific semantics and collaborative signals. OGR uses list-wise preference planning and pipelined SID decoding to model global preferences and inter-item dependencies, further enhanced by SPA, a reward-guided policy optimization method. Experiments show OGR significantly outperforms existing baselines, achieving substantial gains in NDCG and a notable improvement in Effective Views during online A/B testing on Kuaishou. AI
IMPACT This research introduces a novel approach to recommendation systems that could improve user engagement and content discovery in platforms like Kuaishou.
RANK_REASON Academic paper detailing a new recommendation system framework. [lever_c_demoted from research: ic=1 ai=0.7]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
