System-Design-Notes

[System Design] Ads Click Prediction

High-level Design

Ads Features

IDs, Image/Video, Category, Impression and click numbers

User Features

Demographics: age, gender, city, coutry…

Contextual information: device, time…

Interaction-related features: clicked ads, user’s historical engagenment statistics…

Model Selection

• Logistic regression
• Feature crossing + logistic regression
• Gradient boosted decision trees
• Gradient boosted decision trees + logistic regression
• Neural networks
• Deep & Cross networks
• Factorization Machines (FM); (→ FFM)
• Deep Factorization Machines (DeepFM); (→XDeepFM)
• Deep Interest Network (DIN); (→DIEN)

Offline Metrics

• Cross-entropy
• Normalized cross-entropy

Online Metrics

• CTR
• Conversion rate
• Revenue lift
• Hide rate

Summary