Tencent has developed HiGR, a hierarchical generative framework for industrial-scale slate recommendation. This system addresses challenges in applying generative models to large-scale recommendation by learning structured item IDs and shifting autoregressive modeling to preference embeddings for efficient planning. HiGR has demonstrated significant improvements in offline recommendation quality and inference speed, and has been successfully deployed on Tencent platforms, enhancing user engagement metrics. AI
IMPACT This framework could significantly improve recommendation efficiency and effectiveness for platforms serving hundreds of millions of users.
RANK_REASON Academic paper published on arXiv detailing a new framework for slate recommendation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →