E-commercePersonalized semantic retrieval / candidate generation using learned user-item embeddings.production-deployed at large scale; used in homefeed and notifications with auto-retraining and version-synchronized serving.

Two-tower learned retrieval for Pinterest Homefeed candidate generation

Pinterest learns a numerical fingerprint for each user and each Pin, then quickly finds Pins whose fingerprints are closest to the user’s current interests.

10.0
Quality
Score

Executive Brief

Business Problem Solved

Premium content - unlock to see the business problem analysis...

Value Drivers

Value Driver 1Value Driver 2

Strategic Moat

Premium content - unlock to see strategic analysis...

Premium Access

Full Analysis Available

Executive brief, technical architecture, and market positioning for this use case.

Ecommerce — $499/yrAtlas Pro — $1,499/yr
Executive Brief•Technical•Market Signal

Technical Analysis

Model Strategy

Premium content...

Data Strategy

Premium content...

Implementation Complexity

Premium content...

Scalability Bottleneck

Premium content...

Premium Access

Full Analysis Available

Executive brief, technical architecture, and market positioning for this use case.

Ecommerce — $499/yrAtlas Pro — $1,499/yr
Executive Brief•Technical•Market Signal

Market Signal

Adoption Stage

Premium content...

Differentiation Factor

Premium content...

Key Competitors

Company ACompany B
Premium Access

Full Analysis Available

Executive brief, technical architecture, and market positioning for this use case.

Ecommerce — $499/yrAtlas Pro — $1,499/yr
Executive Brief•Technical•Market Signal

Part of Application

Personalized Feed Candidate Retrieval and Search Ranking

2 use cases in this application

Explore More

More in E-commerce→More Personalized semantic retrieval / candidate generation using learned user-item embeddings.→

Source

https://medium.com/pinterest-engineering/establishing-a-large-scale-learned-retrieval-system-at-pinterest-eb0eaf7b92c5