PushDualGen: Enabling LLMs to Generate Semantic IDs with Interpretable Copy for Industrial Push Recommendation¶
Fidelity: core mechanism. Paper results, local public-data measurements, and omissions are separate.
Paper information¶
| Field | Details |
|---|---|
| Paper | arXiv 2608.07989 |
| Organization | Kuaishou Technology |
| First public release | 2026-08-08 (arXiv v1) |
| Authors' code | The paper did not provide official or author code when checked on 2026-08-24. |
| Adapter / method | pushdualgen |
| Local reproduction code | src/auto_research/reproductions/pushdualgen/ |
Original paper¶
Motivation and method¶
Generate a servable semantic ID first, then optional explanatory copy only when needed. Serving fuses SID and copy representations while preserving explanation consistency.
flowchart LR
A["SID compression"] --> B["SID to optional copy"] --> C["Representation fusion"]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¶
The main online result is effective play rate +8.50% over 14 days at 15% traffic (150M users). Offline and online metrics use private or paper-specific protocols. The private paper protocol is not directly comparable with local MovieLens measurements.
Local reproduction¶
Local comparison: The baseline is the shared transition + content scorer. The treatment adds only pushdualgen on the same users, items, full-catalog candidates, and seed.
MovieLens-100K, 220 users and 360 items, seed 42: NDCG@10 0.0540 → 0.0546 (+1.19%); Hit@10 0.1091 → 0.0955. Validation selects mixture weights only; the test set is not used for tuning.
See the fixed metrics.
Reproduction boundaries¶
The core mechanism runs on MovieLens-100K with the same split, fixed full-catalog candidates, and seed. Private features, the production foundation model, and online traffic are unavailable. The implementation has independent model state and scoring, rather than mapping a paper name to a shared weighted formula. See the adapter's omitted_core_components for omissions. Local changes and paper A/B gains are reported separately.