DualGR: Generative Retrieval with Long and Short-Term Interests Modeling¶
Fidelity: core mechanism. Paper results, public-data measurements, and omissions are reported separately.
Paper information¶
| Field | Details |
|---|---|
| Paper | arXiv 2511.12518 |
| Organization | USTC / Kuaishou Technology |
| First public release | 2025-11-16 (arXiv v1) |
| Authors' code | The paper did not provide official or author code when checked on 2026-08-09. |
| Adapter / method | dualgr |
| Local reproduction code | src/auto_research/reproductions/dualgr/ |
Original paper¶
Motivation and method¶
Use separate routers for short- and long-term interests, enforce valid semantic-ID prefixes, and apply an exposure-aware term to prevent concentration on head items.
flowchart LR
A["Public input or history"] --> B["DualGR core path"]
B --> C["Paper-specific state or objective"]
C --> D["Matched-budget evaluation"]Key figure from the paper¶
Figure 1: Overview of the paper's method and key components. From the original paper; copyright remains with the authors.
Core equations¶
Paper results¶
Kuaishou online video views +0.527% and watch time +0.432%.
Local reproduction¶
Local comparison: The baseline is the shared transition + content scorer. Adding only the DualGR core mechanism changes NDCG@10 by +6.73% relative.
MovieLens-100K, 260 users and 420 items, seed 42: NDCG@10 0.0354 → 0.0378 (+6.73%).
auto-research reproduce --paper dualgr --dataset-dir data --seed 42
auto-research evolve --model rankmixer --dataset movielens-100k --direction "组合 dualgr 与已安装论文算子" --generations 2 --population 4
See the fixed metrics.
Reproduction boundaries¶
Full-catalog MovieLens-100K exercises the paper's core state, objective, or inference path. No private features, production traffic, or online serving are used. Paper A/B results are cited separately and local numbers do not represent the original large model, data, traffic, or custom kernels.