A Survey of Recommender Systems and Geographical Recommendation Techniques

Introduction

Nowadays, with the high availability of data, the wide use of social networks and the rapid growth of the internet caused a huge amount of data, it requires a complex process to extract useful information that can be presented to the user in order to help him manage the data properly leading to making correct decisions. Researches have been done that help in managing data to produce this useful information (Al-Otaibi & Ykhlef, 2012). Many users are interested in systems that recommends some products to them based on certain factors, and hence a system is needed in order to support users in selecting a product or any item taking into consideration that the user may have less knowledge about the domain, we call this system a recommender system (RS) (Anderson & Hiralall, 2009). (Ricci, Rokach & Shapira, 2011) define RS as an intelligent system that provides advice to the user about a specific item aiding him in the decision making process (see Figure. 1).

Figure 1.

The model of process for recommender system

Such a system can gain information explicitly (e.g. Collect rating from user) or implicitly (monitoring user behavior, such as a book read or a song heard) (Bobadilla, Ortega, Hernando & Gutiérrez, 2013). With the advancement of location acquisition technologies and fast development of mobile devices and wireless communication, LBSNs have attracted millions of users such as Facebook places, Gowalla, and Foursquare. In LBSNs, users can share their experiences of visiting specific locations, also known as points-of-interests (POIs), for example museums, restaurants, and stores. These visits are also known as check-in activities that reflect the user’s preferences on locations. In LBSNs, it is an important task to make location recommendation to users - it utilizes user preferences and other information (e.g., social friendships), which not only helps users explore new places but also makes LBSNs more attractive to users (Zhang & Chow, 2013). To gain a rich knowledge about users’ interests, we can use available check-in data in LBSNs; it is beneficial to a wide range of applications such as location, activity, and friend's recommendations (Ye, Lee & Lee, 2011). In the LBSNs, the check-in data contains the following unique characteristics:

• •

  Frequency Data and Sparsity: User-location matrix indicates the frequency of a user visiting a place. The density of the data set makes the POI recommendation task very tough.

• •

  Multi-centers and Normal Distribution: Where the check-in locations follow a Gaussian distribution at each center, because users tend to check in around several centers.

• •

  Inverse Distance Rule: The probability of visiting a place is inversely proportional to the distance from its nearest center, although each user has a different personalized taste for POI. This implies that if a place is too far away from the location a user lives in, it is less likely that he would visit that place even if he likes it.

Friendship Influence: This implies that social influence exists; and may have an effect on the check-in activity (Cheng, Yang, King, & Lyu, 2012) (see Figure 2).

Figure 2.

Location-based social network (Zhang & Chow, 2015)