Kuaishou Boosts Recommendation Recall by 9% with Dynamic Single-Level Codebook
_reachsumit · x · 2026-08-24
The paper proposes replacing static multi-level small codebooks with a dynamic single-level large codebook in generative recommendation, reducing decoding steps and adapting to traffic shifts. It introduces an exposure-aware dynamic update mechanism. On public datasets, this method improves OneRec-V1 Recall@10 by 5.0%-8.8% and NDCG@10 by 4.1%-5.1%.
More from Research
- 30-Question Breakdown of LLM Post-Training: SFT, LoRA, RLHF, DPO, GRPO — techNmak · 2026-08-24
- Deep Learning Dimensional Rule: Tensors with Three Large Axes Won't Fly — francoisfleuret · 2026-08-24
- PyTorch Compilation Pitfall: FSDP Guard Skipping Hurts MFU — SeunghyunSEO7 · 2026-08-24
- Tsinghua's OVOW Reconstructs Interactive 4D Physical World from Single Video — jiqizhixin · 2026-08-24
- Research Agent Memory: The Value of Remembering Rejected Experiments — LazyBoy0000 · 2026-08-24
- Cognitive Archeology, Technological Evolution, and the Extended Cognition Agenda — abenitezburraco · 2026-08-24