Research map / Architectures and efficiency / Attention and transformers
CollectiveKV: Decoupling and Sharing Collaborative Information in Sequential Recommendation
Jingyu Li, Zhaocheng Du, Qianhui Zhu, kaiyuan Li, Zhicheng Zhang, Song-Li Wu, Chaolang Li, Pengwen Dai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CollectiveKV cross-user KV sharing | SIM · MicroVideo | AUC: 0.6933 -> 0.7057 | Improved |
| add CollectiveKV | SDIM · MicroVideo | AUC: 0.6749 -> 0.6871 | Improved |
| add CollectiveKV | ETA · KuaiVideo | GAUC: 0.6699 -> 0.6677 | Hurt |
| add CollectiveKV | HSTU · MicroVideo | AUC: 0.7381 -> 0.7385 | No clear effect |
| replace per-user KV cache loading with CollectiveKV at batch size 512 | SIM · KuaiVideo latency | latency (ms): 32.991 -> 0.695 | Improved |
| add peak loss and balance loss to router (ablation) | CollectiveKV · MicroVideo (SIM) | GAUC: 0.6971 -> 0.6973 | Mixed |
| add both router losses (ablation) | CollectiveKV · KuaiVideo (SIM) | GAUC: 0.6604 -> 0.6585 | Hurt |
| share both keys and values instead of only one (ablation) | CollectiveKV · KuaiVideo and EBNeRD | AUC: KuaiVideo 0.6878 -> 0.6900; EBNeRD 0.6991 -> 0.7088 (vs K only) | Improved |
| share both keys and values instead of only values (ablation) | CollectiveKV · MicroVideo (SIM) | AUC: 0.7073 -> 0.7057 | Hurt |
| increase global KV pool size beyond about 7000 (ablation) | CollectiveKV · MicroVideo (SIM) | GAUC/AUC | Hurt |
Closest papers by meaning
- MARS: Multi-resolution Adaptive Routing for Sequential Recommendation (NeurIPS 2026 · similarity 0.88)
- LLM-ESR: Large Language Models Enhancement for Long-tailed Sequential Recommendation (NeurIPS 2024 · similarity 0.87)
- SLMRec: Distilling Large Language Models into Small for Sequential Recommendation (ICLR 2025 · similarity 0.87)
- LIME: Link-based User-item Interaction Modeling with Decoupled XOR Attention for Efficient Test Time Scaling (NeurIPS 2026 · similarity 0.87)
- Augmenting Sequential Recommendation with Balanced Relevance and Diversity (AAAI 2025 · similarity 0.86)
- LLM4RSR: Large Language Models as Data Correctors for Robust Sequential Recommendation (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
