Recommender systems are systems that make recommendations for users on the basis of data the users have entered into the system. The more data a user has provided, the more accurate such systems can be. In addition, data submitted by individual users helps to improve the system overall, by generating information that can be used to make recommendations for other users. Recommender systems are commonly seen on sites like movie and television review sites and those with large inventories of retail items that would be functionally impossible to browse by looking at every item.
These systems can interact with users in a number of different ways. One is as a service to users who are looking for more things they might be interested in, like further reading, television shows, or video games. In these systems, the user generates a list of likes and dislikes and the system tries to predict how the user will vote on things he or she has not voted on yet. If it thinks that something would have a high rating, it suggests it to the user.
Well designed recommender systems learn from their mistakes. A system might recommend The Sound of Music because a user liked Willy Wonka & the Chocolate Factory. The user could select options such as "I like this" or "I don't like this." If the user didn't like The Sound of Music, the system could take note and further refine the algorithm used to generate recommendations. The more data accrued, the more helpful the recommendations will be.
Retail sites use recommender systems to entice people into making impulse purchases. The system takes note of items purchased and recommends related and helpful items. For instance, someone who is buying a camera might be asked if he or she wants to buy a charger, a camera case, filters, and additional lenses. Someone purchasing a book on feminist theory might be told that other buyers of that title also enjoy another, related title. These types of recommender systems allow for personalized marketing that is highly likely to appeal to users.
These systems rely on collaborative filtering of data, in which data from vast numbers of users is organized in meaningful ways. This allows the site to make connections that might not otherwise be apparent, improving the quality of the recommendations. Users who do not want to participate can usually change options in their user settings, but they will reduce the quality of recommendations they receive because the system cannot learn from the preferences of the individual, only the collective opinion of other users.
