RESEARCH · RESEARCH · #1011
Learned Cross-Task Relationships improves multi-task recommendation at YouTube (arXiv:2609.28776v1)
The paper introduces a framework that learns pairwise cross-task relationships to approximate the joint distribution of labels in multi-task models, reducing complexity compared with modeling the full joint space. The authors evaluate the approach in YouTube's production recommendation systems and report accuracy and user-satisfaction improvements across Notifications, Homepage, and Watch Next surfaces, and also provide a workflow template for broader implementation.
KEY POINTS
- The paper introduces a framework that learns pairwise cross-task relationships to approximate the joint distribution of labels in multi-task models, reducing complexity compared with modeling the full joint space.
- The authors evaluate the approach in YouTube's production recommendation systems and report accuracy and user-satisfaction improvements across Notifications, Homepage, and Watch Next surfaces, and also provide a workflow template for broader implementation.
- Learning targeted pairwise task relationships offers a scalable way to transfer information across tasks without modeling a full joint distribution, and the paper demonstrates gains in a major production recommender (YouTube).
WHY IT MATTERS
Learning targeted pairwise task relationships offers a scalable way to transfer information across tasks without modeling a full joint distribution, and the paper demonstrates gains in a major production recommender (YouTube).