MAPS: Netflix’s Multimodal Asset Personalization at Scale

Summary

Netflix treats artwork and video previews as personalized assets, but ID-based models struggle with newly launched titles and assets that have little interaction history. The systems encode artwork with 768-dimensional CLIP image embeddings and combine those vectors with learned asset IDs, allowing visual preferences to transfer across titles and enabling one model to serve five canvases. Reward-based weighting balances pooled canvas data by the long-term value of interactions, while inverse propensity scoring on randomized exploration traffic supports offline evaluation. For video previews, MediaFM combines visual, audio, and text signals and ranked ahead of SeqCLIP in a linear probe, offline IPS evaluation, and online A/B testing; the Netflix Embedding Store supplies consistent embeddings to production models.

Takeaways

1. 1

   CLIP embeddings let artwork personalization transfer preferences across titles and across related crops, so a new asset can receive content-based signals before accumulating its own interaction history.

2. 2

   Reward-based weighting mixes canvas data according to the long-term value of interaction types rather than raw impression volume, avoiding hand-tuned per-canvas weights.

3. 3

   MediaFM outperformed SeqCLIP in linear-probe accuracy, offline IPS lift, and online A/B results; the linear probe is now used to gate new MediaFM versions.