From Understanding to Action: Feedback-Grounded Policy Discovery for Generative Recommendation¶
Core public-data feedback-to-ranking mechanism.
Paper information¶
| Field | Details |
|---|---|
| Paper | Paper or original source |
| Institution | Huazhong Agricultural University (Kuaishou internship). |
| First public release | 2026-07-30 (arXiv v1) |
| Author code | No official author code linked by the paper as checked on 2026-08-24. |
| Local adapter or method key | feedback-policy |
| Local reproduction code | src/auto_research/reproductions/feedback_policy/ |
Paper method and evidence¶
Discover policies from user feedback and distill them into a lightweight ranker using relations in two spaces, avoiding LLM calls at serving.
flowchart LR
A[Discover policy from feedback] --> B[Distill relations in two spaces]
B[Distill relations in two spaces] --> C[Serve lightweight ranker]Original paper figure¶
The original figure shows discovery and two-space distillation. From the original paper; copyright remains with the authors.
Core equation¶
\[
\mathcal L=\mathcal L_{rank}+\lambda\|S_T-S_S\|_F^2
\]
Reported results¶
Large-scale online A/B reports Revenue +4.51%; distinct from MovieLens.
Local reproduction¶
Matched MovieLens-100K 220 users/360 items, full catalog, seed 42: shared transition+content baseline versus feedback policy, NDCG@10 0.0540→0.0555 (+2.76%), Hit@10 0.1091→0.1091. Validation selects blend; no test tuning. Metrics. No private features, production traffic or serving.