Gabriel Benz

These days, the enormous amount of information available on the Internet has been setting for web information systems a new challenge. Namely, a shift in focus from [standard] information access to the way it is presented to the final user.

Given that, recommender systems have been emerging as a solution to this problem. These systems have gained a huge popularity due to the benefits they have provided to companies that use them (e.g. Amazon, Google and Netflix ).

We take advantage of recommendation in day-to-day tasks such as purchasing, software using, appointment scheduling or even to get a reference to some information…

… But what really is a recommender system?

It is a system that helps people to find the information they are looking for by means of indications. In other words, it is a system that strives to predict the user’s preferences based on his/her usage history and/or on the history of other users.

…What can be indicated?

• Advertising messages                                 • Investment choices

• Articles / Quotes                                         • Restaurants / Cafes

• Online mates (Dating services)                   • Friends (Social network sites)

• Courses (e-learning)                                   • TV programs / series

• Products / Musics / Movies / Books / Drugs / Games / Clothes…

In order to understand how these systems work, some techniques and algorithms that are normally used are listed below.

Techniques & Algorithms:

- Top N: The top rated items are recommended (most read, most sold, most downloaded, most indicated, most quoted, etc.).

- Knowledge-based: The rules are defined manually (by editor’s choice), but retrieved by data mining.

- Cases-based: The indication is made by analysing similar cases of other users.

- Content-based: The items the user is suposedly interested in, according to his profile, are recommended. Besides, items that are similar to the recommended are also recommended.

- Collaborative filtering: Uses similarity measuring functions for indicating products. If two users have the same interest, then the same items should be indicated for both.

… In sum, what do recommender systems do, exactly?

1. Calculate how much you may like a certain item.

2. Compose a list of N best items for you.

3. Compose a list of N best users for a certain item.

4. Explain why these items are recommended to you.

5. Adjust the measurement and recommendation based on feedback obtained.

Examples:

Amazon.com: Calculates, among other things, the books and authors more bought by people who bought a certain book and the product rating by customer using Item-to-item collaborative filtering.

Netflix: Uses collaborative filtering to recommend movies analysing the customers’ ratings (rented movies and current queue) and the combined ratings of all Netflix users.

Last.fm: Recommendations are calculated using a collaborative filtering algorithm. Users can hear previews of a list of artists not listed on their own profile but which appear on those of others with similar musical tastes.

MovieLens: Recommends films for its users to watch, based on their film preferences and using collaborative filtering.

Does it really work? (Celma & Lamere, ISMIR 2007)

• Netflix: 66% rented movies are from recommendation.

• Google News: 38% more click-through are due to recommendation.

• Amazon: 35% sales are from recommendation.

To summarize, the main Idea of these systems is to offer the user a customized environment based on his own wills and interests, optimizing the purchasing experience and maximizing the profits.

“If I have 3 million customers on the Web, I should have 3 million stores on the Web.” Jeff Bezos, CEO of Amazon.

“People don’t know what they want until you show it to them.”, Steve Jobs.

“We are leaving the age of information and entering the age of recommendation”, Chris Anderson in The Long Tail.