Recommender systems try to provide people with recommendations of items they will appreciate, based on their past preferences, history of purchase, and demographic information. This chapter (1) introduces recommender systems, classifying them along four dimensions (i.e. the way the preferences are gathered, the used approach, the type of algorithm, and the way the results are provided) and describing recent work done in the area, and (2) provides more details about one such type of recommender systems, namely collaborative-recommendation systems. Such systems work by analyzing the items previously rated by all the users and are not based on the content of the items, as content-based systems.
TopIntroduction
Recommender systems try to provide people with recommendations of items they will appreciate, based on their past preferences, history of purchase, and demographic information. Recommender systems have their origin (see the survey of the state-of-the-art of Adomavicius and Tuzhilin (2005) for more details) in the work done in, mainly, machine learning, information retrieval (Salton, 1989), cognitive science (Rich, 1979), forecasting theories (Armstrong, 2001), marketing (Lilien, Smith & Moorthy, 1992), management (Murthi & Sarkar, 2003), and emerged as an independent research area in the mid-1990s, with the first papers on collaborative filtering (Hill, Stead, Rosenstein & Furnas, 1995; Resnick, Neophytos, Mitesh, Bergstrom & Riedl, 1994; Shardanand & Maes, 1995).
Three steps usually are common to the functioning of recommender systems:
- 1.
Gather valuable information on the users (past preferences, demographic information, etc.) and on the items (description, keywords, etc.).
- 2.
Determine patterns from these historical data.
- 3.
Suggest items to people.
TopTypology Of Recommender Systems
Many different ways have been developed to achieve the final goal of making recommendations to persons. This section reviews the main features of recommender systems:
- 1.
Related to the first step, the way the preferences are gathered is described in the Section “Preference indicators”.
- 2.
Related to the second step, various ways developed to extract new information from data are analyzed. Finding new information depends on two axes, the first one describing the global approach (content-based or collaborative approaches in the Section “Filtering approach”), while the second one describes the type of algorithm (memory-based or model-based algorithms in the Section “Recommendation algorithm”).
- 3.
Related to the third step, the two possibilities existing for providing results to a user are introduced in the Section “Prediction or recommendation”.