Like the user-user approach, items can be used to find similarities and then items can be recommended. The idea is to locate similar items based on who is purchasing them. This approach is generally taken when user bases grow. So, if the number of users is denoted by M and number of items is denoted by N, when M >> N (read: M is much greater than N) then this approach yields a good result. Most of the e-commerce sites use this algorithm to recommend items to users.
Amazon's "People who bought this also bought these" is a simple output of this approach. You can think of this as an inverted index of items and users in a very simple way, devoid of all the details about how ratings are persisted.
Items |
Users |
---|---|
I1 |
U1,U2,U3,U5 |
I2 |
U3,U1 |
I3 |
U1,U5 |
Now, if a new user buys item I2, then the system will list all the other items that have been purchased by U3 and U1, which are I1 and I3. On the other hand, if the new user buys item I3, then the system will...