CADET: Context-Conditioned Ads CTR Prediction With a Decoder-Only Transformer¶

Core public-data mechanism, not the production A/B.

Paper information¶

Field	Details
Paper	arXiv 2602.11410
Institution	LinkedIn.
First public release	2026-02-11 (arXiv v1)
Author code	No official author code linked by the paper as checked on 2026-08-24.
Local adapter or method key	`cadet`
Local reproduction code	`src/auto_research/reproductions/cadet/`

Paper method and evidence¶

A decoder-only Transformer encodes cacheable histories, then injects post-scoring context after candidate scoring to align training with serving and unify ad CTR prediction.

flowchart LR
  H[Cacheable history] --> E[Decoder-only encoder]
  E --> S[Candidate scoring]
  C[Post-scoring context] --> A[Conditioned attention]
  S --> A

Original paper figure¶

The original figure shows cacheable history and post-scoring context. From the original paper; copyright remains with the authors.

Core equation¶

\[ h_i'=h_i+\operatorname{Attn}(q(c),K(h_{\le i}),V(h_{\le i})) \]

Reported results¶

Section 5.3/Table 3 reports CTR +11.04% in large-scale online A/B.

Local reproduction¶

NDCG@10 0.0540→0.0507 (-6.20%), Hit@10 0.1091→0.1000. Same MovieLens-100K 220 users/360 items, full catalog, split and seed 42, using a shared transition + content baseline. Validation selects only blend weight; test does not tune. The local model has independent state and scoring. No private features, production checkpoint or online serving; retain negative results.

auto-research reproduce --paper cadet --dataset-dir data --seed 42

Fixed metrics.