Article Preview
Top1. Introduction
A recommender system (RS) is data-driven software that helps its users to find relevant items in an overloaded search space (Schafer, Konstan and Riedl 2001; Burke 2002). Providing each user with personalized information gives such system an intelligent character. So, the more satisfactory the items provided to the user, the more efficient the system. Recommendation is present in several areas of our daily life: e-commerce (Schafer, Konstan and Riedl 2001; Resnick et al. 1994; Huang, Zeng, Chen 2007; Chen and Tai 2004; Jinghua, Kangning, Shaohong 2007), e-learning (Yera Toledo and Caballero Mota 2014; Bobadilla, Serradilla and Bernal 2010; Denis 2007; Zaiane 2002), tourism (Noguera et al. 2012), libraries (Yang and Lin 2013), scientific articles (Bai et al. 2019), music (Lee, Cho and Kim 2010; Nanolopoulus et al. 2010). The most common well studied field is movie recommendation (Antonopoulus and Salter 2006; Konstan, Miller and Riedl 2004; Li and Yamada 2004; Carrer-Neto et al. 2012; Winoto and Tang 2010).
The classification of recommender systems depends on information used to generate recommendations. Basically, there are two classes which are: Content-based recommendation (Gemmis et al. 2015), Collaborative Filtering based recommendation (Ning, Desrosiers, and Karypis 2015). The former uses features of items and user’s behavior to generate recommendations (Lang 1995), i.e. a content-based movies recommender uses genre of movies (action, comedy, horror, romantic) and movies preferred by a target user in the past to generate recommendations (predicting relevant movies). The second recommends to the target user the items that satisfied similar users. Similarity could be computed using rating information (Ekstrand, Riedl and Konstan 2011; Resnick et al. 1994); two users are similar if they have rated (voted) items in the same way. Several metrics exist in literature to measure similarity between users like Pearson correlation which remains a reference (Ingoo, Kyong and Tae 2003; Adomavicius 2005; Bobadilla, Serradilla and Hernando 2009). Hybridization uses the two previous techniques to generate recommendations (Burke 2002; Bobadilla et al. 2013). Moreover, demographic information could be used in the recommendation process, i.e. age, sex, nationality. Such technique supposes that users with common demographic information will have common preferences (Krulwich 1997; Burke 2002).