RecommendationSystem-[TODO]

• Recommendation system

  • Use cases

  • Components

  • Overview

    • Cold start

    • Offline processing

    • Online processing

  • Design

    • Version 1

    • Version 2

    • Version 3

Recommendation system

Use cases

• Long tail

• Utilize the traffic

• Same or close categories of products

  • Collaborative filtering

    • e.g. The number of people viewing A 10, the number of people viewing B 20. The number of people viewing A and B is 5. Then the correlation coefficient between A and B is "A intersect B" / "A union B"

Components

• Regression

• Sort

Overview

Cold start

• Recommend something popular

  1. Based on the content / based on the user behavior

  2. Compute the correlation coefficient

  3. Sort

Offline processing

Online processing

Design

Version 1

• No user customization

• Factors

  • Number of pictures

  • The length of online

• Fixed weights

• Making sure that not all results come from a single saler

• No support for online user testing

Version 2

• Overtime users have accumulated their behavior

• Record users' behavior into Hadoop cluster

Version 3