Within the field of collaborative filtering, we can usefully make the distinction between two types of filtering—item-based and user-based recommenders. With item-based recommenders, we take a set of items that a user has already rated highly and look for other items that are similar. The process is visualized in the next diagram:
A recommender might recommend item B, based on the information presented in the diagram, since it's similar to two items that are already highly rated.
We can contrast this approach to the process of a user-based recommendation shown in the following diagram. A user-based recommendation aims to identify users with similar tastes to the user in question to recommend items that they have rated highly, but which the user has not already rated.
The user-based recommender is likely to recommend item B, because it has been rated highly by two other users with similar taste. We'll be implementing both kinds of recommenders in this chapter...