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

2026-06-04 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

Collaborative filtering and graph-based recommendation models are highly effective because they leverage observed user interactions, but this dependence creates a fundamental cold-start challenge when newly added content has no interaction history. Tubi's production retrieval system addresses this by formulating cold-start recommendation as an inductive graph-completion problem on a temporal bipartite device-content graph. They propose Shallow-RHS, an asymmetric link-prediction architecture where the left-hand side (LHS) device tower uses watch-history message passing for collaborative signals, while the right-hand side (RHS) content tower is intentionally shallow, encoding content solely from intrinsic features. This forces the content encoder to map intrinsic features into a collaborative-filtering-aware embedding space. After training, the learned content encoder generates embeddings for both warm and newly ingested content, enabling implicit graph completion. The principle extends to device cold-start using cohort-based demographic embeddings. Large-scale online experiments demonstrated consistent relative improvements in content cold-start engagement, promotion speed, impression acquisition, and device cold-start engagement.

Key takeaway

For Machine Learning Engineers building recommendation systems, if you are struggling with cold-start content or device recommendations, consider implementing an asymmetric graph architecture like Shallow-RHS. This approach allows you to generate immediate, collaborative-filtering-aware embeddings for new items using only intrinsic features, significantly improving engagement and promotion speed. You can also extend this principle to new users via cohort-based demographic embeddings.

Key insights

Shallow-RHS uses an asymmetric graph architecture to generate cold-start item embeddings from intrinsic features, bridging semantic and collaborative signals.

Principles

• Asymmetric graph architectures can address cold-start challenges.
• Intrinsic features can be mapped to collaborative-filtering-aware spaces.
• Inductive graph completion handles new entities effectively.

Method

Formulate cold-start as inductive graph completion on a temporal bipartite device-content graph. Use an asymmetric link-prediction architecture with a deep LHS device tower and a shallow RHS content tower encoding intrinsic features.

In practice

• Generate immediate embeddings for new content.
• Enable approximate nearest-neighbor retrieval for cold-start.
• Construct cohort-based device embeddings from demographics.

Topics

• Cold-Start Recommendation
• Graph Neural Networks
• Collaborative Filtering
• Item Embedding
• Inductive Graph Completion
• Tubi

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• Bridging the Semantic-Collaborative Gap: An Asymmetric Graph Architecture for Cold-Start Item Recommendation — An asymmetric graph architecture can bridge the semantic-collaborative gap for cold-start recommendations by aligning intrinsic features with behavioral signals.
• Knowledge Graph and Hypergraph Transformers with Repository-Attention and Journey-Based Role Transport — A novel architecture separates structured knowledge from language in transformers using a key-value repository and journey-based role transport.
• How Embeddings Shape Graph Neural Networks: Classical vs Quantum-Oriented Node Representations — Quantum-oriented graph embeddings offer consistent gains on structure-driven datasets compared to classical methods.
• How Recommendation System Works on Youtube — YouTube's recommendation system uses a two-stage deep learning approach to personalize content at massive scale.
• Cracking the Cold Start Problem — Hybrid recommender systems combining collaborative filtering, embeddings, and bandit learning improve cold start and fairness.
• SilverTorch: Index as Model — A New Retrieval Paradigm for Recommendation Systems — Unifying recommendation retrieval into a single neural network significantly boosts efficiency, quality, and development velocity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.