HCUB: hierarchical contextual uplift bandit¶

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 2601.14333
Organization	Dream11
First public release	2026-01-20 (arXiv v1)
Authors' code	No public author code was found when checked on 2026-09-05.
Adapter / method	`hcub`
Local reproduction code	`src/auto_research/reproductions/hcub/`

Original paper¶

Motivation and method¶

Choose granularity dynamically in a user and system context tree. Sparse child nodes inherit reward from parents to improve cold starts. The policy directly estimates action uplift instead of just fitting observed absolute clicks.

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

Core equations¶

\[ a_t=\arg\max_a\hat\tau(a\mid c_t)+\beta U(a,c_t),\qquad R(c)=\eta R_{local}+(1-\eta)R_{parent}. \]

Paper results¶

Dream11 reports online revenue +0.51%, engagement +0.42%, and about 4%–5% lower offline regret. This is paper evidence, not a local public-data outcome.

Local reproduction¶

Local comparison: Under the same full-catalog MovieLens protocol, baseline NDCG@10 is 0.05401 and treatment NDCG@10 is 0.04886, -9.53% relative. The local proxy and production task differ, so there is no extrapolation to online lift.

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

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

Reproduction boundaries¶

The local run uses a fixed public MovieLens 100K subset and auditable proxies to check the core computation only. It does not reproduce private logs, the production foundation model, online traffic assignment, or serving. There is no claim to reproduce absolute paper metrics or online gains.