OneRank: A native unified Transformer for multi-task ranking¶
Core public-data ranking mechanism.
Paper information¶
| Field | Details |
|---|---|
| Paper | Paper or original source |
| Institution | Renmin University of China. |
| First public release | 2026-06-15 (arXiv v1) |
| Author code | No official author code linked by the paper as checked on 2026-09-02. |
| Local adapter or method key | onerank |
| Local reproduction code | src/auto_research/reproductions/onerank/ |
Paper method and evidence¶
Put structured features and task tokens in one sequence instead of separating encoder and task predictors. Task-private channels and isolated gradients limit interference.
flowchart LR
A[Structured features and task tokens] --> B[Unified Transformer]
B[Unified Transformer] --> C[Private channels and gradient isolation]Original paper figure¶
Figure shows unified tokens and task-private paths. From the original paper; copyright remains with the authors.
Core equation¶
\[
h_k=\operatorname{Transformer}([x_1,\ldots,x_n,t_k];\theta,\phi_k),\quad \nabla_{\phi_j}L_k=0\\;(j\ne k).
\]
Reported results¶
Shopee seven days, 10%/10% arms: GMV/UU +1.01%, paid GMV/UU +1.17%, ad revenue/UU +0.81%, bad-query rate -2.29%.
Local reproduction¶
MovieLens-1M full catalog: NDCG@10 0.05401→0.09540 (+76.65%). Three-seed metrics, mean, SD, and 95% CI. No private logs, production models, bucketing or serving. Public proxies cannot establish online lift.