Recommender Systems: What they are and how they make money (1)?

Every day, approximately 20 million words of technical information are recorded. A reader capable of reading 1000 words per minute would require 1.5 months, reading eight hours every day, to get through one day's output, and at the end of that period would have fallen 5.5 years behind in his reading" [1]

There has been an exponential increase in the volume of available digital information (e.g. videos in Youtube and Netix, music in LastFm, electronic resources (e.g. research papers in CiteULike), and online services (e.g. Flicker, Delicious, Amazon) in recent years. This information overload has created a potential problem, which is how to filter and efficiently deliver relevant information to a user.

Information Overload is a problem

Furthermore, information needs to be prioritised for a user rather than just filtering the right information; otherwise, it could become overwhelming. Search engines help Internet users by filtering pages to match explicit queries, but it is very di cult to specify what a user wants by using simple keywords. The Semantic Web also provides some help to find useful information by allowing intelligent search queries; however, it depends on the extent to which the web pages are annotated. These problems highlight a need for information filtering systems that can lter unseen information and can predict whether a user would like a given resource. Such systems are called recommender systems, and they mitigate the forementioned problems to a great extent. 

Recommender systems are information filtering systems, which suggest interesting resources (i.e. movies, books, music, people, etc.) to users based on their preferences what they like or dislike about a particular resource with the goal that these resources are likely to be of interest to users. They process the historical data about users' preferences using machine learning algorithms and learn a model that can compile a ranked list of all resources available for recommendation for each user based on the information encoded in their problem. The highly ranked resources are then recommended to the corresponding user based on the rationale that these resources are most likely to be consumed next by this user. 

Recommender Systems: Customer to Products

Nowadays, a number of recommender systems have been built that help people to find useful resources, spanning a number of areas such as movies (MovieLens, Netix, FilmTrust, etc.); music (CDNOW, Ringo, LastFm, etc.); pictures (Flicker); e-commerce (Amazon, Ebay, etc.); expertise finder (ReferralWeb, Linkedin, etc.); news filtering (Google news); books (whichbook.net); and holidays and travel (tripadvisor.co.uk).

Recommender systems are now considered a salient part of any modern e-commerce system because they help increase the e-commerce systems sales by making useful recommendations items a customer/user would be most likely to consume. The statement, given by Greg Linden, who implemented the first recommendation system for Amazon, shows how the recommender systems help industry to make products: 

(Amazon.com) recommendations generated a couple orders of magnitudemore sales than just showing top sellers"

Money Added to Amazon by Recommender Systems

In the next post, I would briefly describe what sort of data mining and machine learning algorithms are used for generating recommendations. I would describe what recommendation algorithms, the Amazon is using for making money.