IntSR: An Integrated Generative Framework for Search and Recommendation¶
Public full-catalog core mechanism, not production serving.
Paper information¶
| Field | Details |
|---|---|
| Paper | Paper or original source |
| Institution | Alibaba / Amap. |
| First public release | 2025-09-25 (arXiv v1) |
| Author code | No official author code linked by the paper as checked on 2026-08-09. |
| Local adapter or method key | intsr |
| Local reproduction code | src/auto_research/reproductions/intsr/ |
Paper method and evidence¶
Use explicit queries and implicit session intent to generate POIs/items from one search-and-recommendation vocabulary. Temporal context represents intent drift.
flowchart LR
A[Query and session intent] --> B[Time-conditioned generator]
B[Time-conditioned generator] --> C[Unified POI and item vocabulary]Original paper figure¶
Figure shows unified search and recommendation generation. From the original paper; copyright remains with the authors.
Core equation¶
\[
p(i|q,h,t)=\sum_zp(i|z,t)p(z|q,h),\quad\mathcal L=-\log p(i^+|q,h,t).
\]
Reported results¶
Amap reports GMV +9.34% and POI CTR +2.76%.
Local reproduction¶
MovieLens-100K, 260 users/420 items, seed 42: shared transition+content baseline vs IntSR NDCG@10 0.0354→0.0473 (+33.56%).
auto-research reproduce --paper intsr --dataset-dir data --seed 42
auto-research evolve --model rankmixer --dataset movielens-100k --direction "Combine intsr with installed paper operators" --generations 2 --population 4
Fixed metrics. No private features, production traffic or serving. These local public-data results do not reproduce online lift.