#embedding-learning News & Analysis

Bridging the Semantic-Collaborative Gap: An Asymmetric Graph Architecture for Cold-Start Item Recommendation

Researchers at Tubi have developed Shallow-RHS, a graph-based recommendation system that addresses the cold-start problem for new content by using asymmetric neural architectures. The model separates user-interaction modeling from content feature encoding, enabling immediate embeddings for newly ingested items while maintaining collaborative filtering capabilities in production environments.