- 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"
- Collaborative filtering
- Regression
- Sort
- Recommend something popular
- Based on the content / based on the user behavior
- Compute the correlation coefficient
- Sort
- 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
- Overtime users have accumulated their behavior
- Record users' behavior into Hadoop cluster