Chapter Preview
TopIntroduction
It is increasingly difficult to find the right information on the Web in the age of explosive information overload. Recommender systems provide users with personalized suggestions of goods, services, or information and thus help them find the most relevant and interesting goods, services, or information for them. Over the last two decades since the first major recommender systems emerged in the mid-1990s (Konstan et al., 1997; Resnick & Varian, 1997), numerous recommender systems have been developed and used in various application domains including e-commerce, education, and engineering (Aggarwal, 2016; Jannach, Zanker, Felfernig, & Friedrich, 2011; Manouselis, Drachsler, Verbert, & Santos, 2014; Ricci, Rokach, & Shapira, 2015; Robillard, Maalej, Walker, & Zimmermann, 2014). Recommender systems have also proven very useful in various application domains.
A basic personalized recommender system suggests a list of items that seem to be most relevant for a given single target user without considering the context that the user is in by using users’ ratings of items on a single overall criterion where both users and items are in a single domain (Jannach et al., 2011). The basic recommender system can be extended in several ways. There are four major extensions, i.e., suggesting items for a group of target users rather than a single user (group recommendations), suggesting items by considering a specific context of the target user (context-aware recommendations), suggesting items using ratings on multiple criteria rather than a single overall criterion (multi-criteria recommendations), and suggesting items by using users and items in multiple domains rather than a single domain (multiple-domain recommendations).
In this chapter, we present a brief and systematic overview of four major advanced recommender systems — group recommender systems, context-aware recommender systems, multi-criteria recommender systems, and cross-domain recommender systems. We characterize and compare them within a unifying model as extensions of the basic recommender systems. Future research topics and directions in the area of advanced personalized recommendations are discussed.
Key Terms in this Chapter
Group Recommender System: A recommender system that provides a group of users as a whole with a shared list of items that are most relevant to the users in the group.
Multi-Criteria Recommender System: A recommender system that provides a target user with a list of items that are most relevant to the target user by using the relevance ratings of items in multiple criteria that are provided by the users.
Advanced Recommendation: Personalized recommendation as an extension of basic recommendation such as group recommender systems, context-aware recommender systems, multi-criteria recommender systems, and cross-domain recommender systems.
Recommender system: A software system that provides a single target user within a single context with personalized recommendations of items such as goods, services or information to guide the target user to find most relevant items using ratings on a single relevance criterion (i.e., overall) and where both users and items are in a single domain.
Context-Aware Recommender System: A recommender system that provides a target user within a specific context with a list of items that are most relevant to the target user in the specific context.
Cross-Domain Recommender System: A recommender system that provides a target user with a list of items in the target domain that are most relevant to the target user by exploiting knowledge from the source domain that shares resources with the target domain.
Collaborative Recommender System: A recommender system that recommends items through user collaborations and are the most widely used and proven method of providing personalized recommendations. There are three major types: user-to-user collaborative filtering based on user-to-user similarity, item-to-item collaborative filtering based on item-to-item similarity, and latent factor model-based collaborative filtering based on user-item matrix factorization.
Basic Personalized Recommendation: Personalized recommendation of items for a single target user in a single context by using users’ ratings of items on a single overall relevance criterion where both users and items are in a single domain.