Multimedia Asset Personalization via Multimodal Embeddings at Netflix¶

Fidelity: core mechanism. Paper results, local public-data measurements, and omissions are separate.

Paper information¶

Field	Details
Paper	arXiv 2608.18322
Organization	Netflix
First public release	2026-08-18 (arXiv v1)
Authors' code	The paper did not provide official or author code when checked on 2026-08-24.
Adapter / method	`netflix-mediafm`
Local reproduction code	`src/auto_research/reproductions/netflix_mediafm/`

Original paper¶

Motivation and method¶

Use frozen CLIP/MediaFM representations as shared online features in an asset two-tower model and enhance the search canvas with query similarity. Offline IPS and a winner proxy screen checkpoints.

flowchart LR
 A["Frozen multimodal representations"] --> B["Unified asset two-tower model"] --> C["Query-aware ranking"]

Key figure from the paper¶

Figure 1: Core architecture, modules, and connections. From the original paper; copyright remains with the authors.

Core equations¶

\[ s(a|u,q)=\alpha\langle u,a\rangle+(1-\alpha)\langle q,a\rangle \]

Paper results¶

Search canvas A/B reports playthrough rate +0.36%. Offline tables and online metrics use private or paper-specific protocols. The private paper protocol is not directly comparable with local MovieLens measurements.

Local reproduction¶

Local comparison: The baseline is the shared transition + content scorer. The treatment adds only netflix-mediafm on the same users, items, full-catalog candidates, and seed.

MovieLens-100K, 220 users and 360 items, seed 42: NDCG@10 0.0540 → 0.0489 (-9.52%); Hit@10 0.1091 → 0.1000. Validation selects mixture weights only; the test set is not used for tuning.

auto-research reproduce --paper netflix-mediafm --dataset-dir data --seed 42

See the fixed metrics.

Reproduction boundaries¶

The core mechanism runs on MovieLens-100K with the same split, fixed full-catalog candidates, and seed. Private features, the production foundation model, and online traffic are unavailable. The implementation has independent model state and scoring, rather than mapping a paper name to a shared weighted formula. See the adapter's omitted_core_components for omissions. Local changes and paper A/B gains are reported separately.