Transformer model improves personalized recommendations on Disney plus
DRanker: A Transformer-Based Multi-Task Ranking Model for the Disney+ Homepage
Machine Learning
Summary
Streaming platforms often need to decide what content to show you so that you enjoy using the service. The researchers built DRanker, a computer model that analyzes your past viewing habits using a type of AI called transformers. This model ranks shows and movies to decide what to recommend on the Disney+ homepage. It also uses a method to remove biases from recommendations. Using DRanker, Disney+ saw small but meaningful improvements in user engagement and active users worldwide.
What this means in practice
- •For streaming platform engineers: Implement transformer-based ranking models that process user engagement histories to boost homepage recommendation quality and user engagement.
- •For ecommerce platform developers: Use transformer encoders with debiasing techniques to rank product recommendations from user interaction sequences, enhancing personalized shopping experiences.
Authors
Priya Nirmal Singh Khokher, Daniel Nemirovsky, Benjamin P. Whitesell, Ethan Sukrae Lee, Eric Wulf, Tom Chee-Duffy, Nicholas Flores
Abstract
We present DRanker (Disney+ Ranker), a transformer-based multitask ranking model serving personalized recommendations on the Disney+ homepage. We describe the encoder-only architecture over enriched engagement-history sequences paired with a lightweight causal debiasing mechanism, and the inference stack that enables transformer-based ranking at scale. Through iterative development, we systematically evaluated more complex alternatives and found that simpler designs consistently outperformed them on offline metrics, qualitative behavioral evaluation, and in production. Deployed globally, DRanker drives over 1% lift in our key engagement metric and 0.5% lift in daily active users across all experiences powered by the ranker. We share practical lessons on data engineering, debiasing, and architectural choices that shaped our ranking system at scale.