SID-Coord: coordinated ranking with semantic IDs¶

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 2604.10471
Organization	Kuaishou Technology
First public release	2026-04-12 (arXiv v1)
Authors' code	No public author code was found on 2026-09-05.
Adapter / method	`sid-coord`
Local reproduction code	`src/auto_research/reproductions/sid_coord/`

Original paper¶

Motivation and method¶

Rather than treating semantics as another dense feature, build multilevel discrete SIDs. Hierarchical fusion, target-aware HID–SID gating, and historical-interest alignment coordinate head-item memorization and long-tail generalization per example.

flowchart LR
  A["Public behavior and candidates"] --> B["SID-Coord 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 3: Overview of the paper's method and key components. From the original paper; copyright remains with the authors.

Core equations¶

\[ s_i=(1-g_i)s_{HID,i}+g_i\sum_l\alpha_l s_{SID_i^l}+\lambda\,\operatorname{Align}(u,i). \]

Paper results¶

Short-video search online A/B reports long-play rate +0.664% and play time +0.369%. 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 SID-Coord NDCG@10 is 0.04970, -7.97% 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 sid-coord --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.