GenPage: Towards End-to-End Generative Homepage Construction at Netflix

2026-06-29

AI Summary

This work proposes the first end-to-end generative homepage recommendation framework, addressing the high complexity and latency of traditional multi-stage recommender systems that hinder holistic user experience optimization. Built upon a single Transformer model, the framework autoregressively generates structured multi-row layouts conditioned on user and request context as prompts. It innovatively integrates a post-training strategy combining weighted binary classification with reinforcement learning to embed business rule constraints, while introducing an efficient online serving architecture to tackle industrial challenges such as cold-start scenarios and deployment efficiency. Online A/B tests demonstrate a statistically significant 0.24% (p<0.001) improvement in core user engagement and a 20% reduction in end-to-end latency. Offline experiments further confirm that prompt augmentation outperforms mere model scaling, and that reinforcement learning substantially enhances recommendation diversity.

Abstract

We present GenPage, an end-to-end generative approach to Netflix homepage construction that replaces the traditional multi-stage recommender stack with a single transformer. GenPage treats the user and request context as a prompt, and autoregressively generates the entire structured, multi-row homepage as the response. We adapt the LLM training recipe: pretraining on production pages, followed by post-training via weighted binary classification (WBC) or reinforcement learning (RL). For industry-scale deployment, we introduce techniques addressing cold start, model freshness, business-rule enforcement, and serving efficiency. In online A/B tests against a mature, highly optimized production homepage recommender, the WBC variant of GenPage delivered a +0.24% lift on the core user engagement metric we use for launch decisions (p < 0.001), while reducing end-to-end serving latency by 20%. Offline, two findings stand out: enriching the prompt yields a larger improvement than scaling model capacity in our current regime, and RL post-training increases homepage diversity even though diversity is not part of the objective.

Problem

• generative homepage construction
• end-to-end recommendation
• transformer-based generation
• user engagement
• homepage personalization

Innovation

• end-to-end generative recommendation
• transformer-based homepage generation
• prompt-based autoregressive modeling
• weighted binary classification
• reinforcement learning for diversity