AIR: atomic intent reasoning for cross-domain recommendation¶

简体中文

Fidelity: core mechanism. Runs the defining mechanism verifiable on public data. Private data, production models, and serving remain outside scope.

Paper information¶

Field	Details
Paper	arXiv 2606.10357
Organization	The Hong Kong Polytechnic University
First public release	2026-06-09 (arXiv v1)
Authors' code	No public author code was found on 2026-09-02.
Adapter / method	`atomic-intent`
Local reproduction code	`src/auto_research/reproductions/atomic_intent/`

Original paper¶

Motivation and method¶

Move expensive LLM reasoning offline and convert individual behaviors into cacheable hierarchical atomic intents. At serving time retrieve intent chains for the target product and aggregate cross-domain evidence with multi-head attention.

flowchart LR
  A["Public behavior and candidates"] --> B["AIR core mechanism"]
  B --> C["Matched-budget method output"]
  A --> D["Common direct baseline"]
  C --> E["Full-catalog ranking and structural diagnostics"]
  D --> E

Key figure from the paper¶

Figure 2: Overall workflow, key stages, and data flow. From the original paper; copyright remains with the authors.

Core equations¶

\[ w_u=\\log(1+N(u))\\sum_k e^{-\\lambda\\Delta t_k},\\qquad \\operatorname{score}(u;v)=\\cos(e_u,e_v)\\log(1+w_u). \]

Paper results¶

Kuaishou e-commerce reports GMV +3.446% on 5.08% of production traffic, and about 400× higher throughput for offline intent caching versus real-time LLM calls. This is paper evidence, not a local public-data outcome.

Local reproduction¶

Local comparison: Under the same full-catalog protocol, baseline NDCG@10 is 0.05401 and AIR NDCG@10 is 0.05240, -2.98% relative. The local proxy does not extrapolate to online lift.

See the three-seed artifact for complete results, means, standard deviations, and 95% confidence intervals.

auto-research reproduce --paper atomic-intent --dataset-dir data --seeds 42,43,44

Reproduction boundaries¶

The local run uses a public MovieLens-1M subset and auditable proxies to validate the core computation. It does not reproduce private logs, production models, online traffic assignment, or serving. No claim is made to reproduce absolute paper metrics or online gains.