Recommendation systems are information filtering systems whose goal is to provide its users with useful recommendations. To determine these recommendations, a recommendation system can use historical data about the user's activity, or it can use recommendations that other users liked (for more information, refer to "A Taxonomy of Recommender Agents on the Internet"). These two approaches are the basis of the two types of algorithms used by recommendation systems—content-based filtering and collaborative filtering. Interestingly, some recommendation systems even use a combination of these two techniques to provide users with recommendations. Both these techniques aim to recommend items, or domain objects that are managed or exchanged by user-centric applications, to its users. Such applications include several websites that provide users with online content and information, such as online shopping and media.

In content-based filtering, recommendations are determined by finding similar items by using a particular user's rating. Each item is represented as a set of discrete features or characteristics, and each item is also rated by several users. Thus, for each user, we have several sets of input variables to represent the characteristics of each item and a set of output variables that represent the user's rating for the item. This information can be used to recommend items with similar features or characteristics as items that were previously rated by a user.

Collaborative filtering methods are based on collecting data about a given user's behavior, activities, or preferences and using this information to recommend items to users. The recommendation is based on how similar a user's behavior is to that of other users. In effect, a user's recommendations are based on her past behavior as well as decisions made by other users in the system. A collaborative filtering technique will use the preferences of similar users to determine the features of all available items in the system, and then it will recommend items with similar features as the items that a given set of users are observed to like.