Recommendation System

Tai Wei

Table of Contents

  • Bullet One
  • Bullet Two
  • Bullet Three
  • Architecture
  • Feature Store (數據中台)
  • Candidate Layer
  • Ranking Layer
  • Reinforcement Learning

Architecture

Candidate Layer

Candidate Layer

  • Inverted-index according to Tag-based feature
  • CF
  • Followers + subscribed topics

Ranking Layer

Ranking Layer

  • Model of YouTube
  • Harmonic Mean of Time + Like
  • According to CTR

Ranking Layer: Model of YouTube

Reinforcement Learning

  • Use Multi-armed Bandit
  • Reward: engagement of articles
  • Action: Choose which model

Multi-armed Bandit

  • Explore
  • Exploit

Architecture

Architecture

推薦系統

User Requests

Feature Store (中台)

Made with Slides.com