Information and Recommender Systems

2Recommender Systems

Nous avons une superbe aiguillette de boeuf mode en gelée. Je me permettrai de vous la recommander. Cest très bien. A moins que vous nayez envie du poulet à l'estragon, qui est très bien aussi (We have a superb aiguillette of jellied beef, which I would recommend. It’s very good. Unless you would prefer our chicken with tarragon, which is also very good)

Jean Dutourd, The Horrors of Love, 1963

A recommender system is an information-filtering technique used to present the items of information (video, music, books, images, Websites, etc.) that may be of interest to the user (Figure 2.1).

Figure 2.1. The recommender system seen as a black box [JAN 10]

2.1. Introduction

Recommender systems have been studied in the context of a range of domains, including information retrieval [KEN 71, SAL 83, BAE 99], the Internet [BAE 04, WHI 07], e-commerce [SCH 01], Web usage mining [SRI 00, FU 02, PIE 03, BAE 05] and many others. The key problem addressed by recommendation may be summarized as an estimation of scores for items that have not yet been seen by a user. The number of items and the number of system users may be very high, making it difficult for every user to view every item or for each item to be evaluated by all users. A method is, therefore, required to estimate the scores for non-viewed items.

Intuitively, this evaluation is generally based on the scores ...