Recommender systems are a part of any internet user’s life now; these systems have evolved together with the Internet, were developed to filter the information better and more precisely. This paper will address ten sources that refer to the discussed topic. Modern approaches, e.g. user-based approach or hybrid recommendation approaches, were described by Asanov (2016); the author examines challenges and issues that modern recommendation systems face.
Anna B. (2016) argues that it is not the accuracy that is important to recommender systems but rather the multiplicity that many performance metrics bring (diversity, coverage, novelty, etc.); there is a chance that an accurate prediction will simply suggest those items the user already has, while metrics provide a more diverse ranking. Bobadilla, Ortega, Hernando, and Gutiérrez (2013) review current recommender systems and presume that these systems will use local and personal data of the users; the authors also approach filtering methods that the systems use, categorize them, and explain their evolution and existing hybrid techniques.
Algorithms criteria of recommender systems, i.e. ‘quality of predictions’ (better than an average RS), ‘speed and scalability’ (the RS has not only to work fast but also to analyze significant amounts of data correctly), and ‘easy update’ (the updates must be handled quickly) is described on the website cs.carleton.edu (2016). One of the most famous recommender systems today, the Netflix Recommender System, is discussed by Gomez-Uribe and Hunt (2015).
They describe various algorithms (e.g. personalized video ranker, trending now, page generation) that Netflix uses, explain the business purpose of the RS, and also present the current innovations in the recommender systems. To better understand the world of the RS and explain its importance, various scientists from such fields as data mining, decision support systems, statistics, and others present theories, trends, challenges of the modern recommender systems in the book edited by Ricci, Rokach, Shapira, and Kantor (2011). Algorithmic approaches that cover buying proposals for users (content-based filtering), the effectiveness of the systems, and practical case studies are presented by Jannach (2011); the author also discusses the recommender systems in social networks as well as consumers’ behavior and the theory linked to it.
Another introduction to recommender systems, the processes behind them, the basic and the hybrid approaches of the systems is described by Jones (2013); cluster algorithms (they provide structure in a seemingly random data) are also approached. Jones (2013) also addresses algorithms and their variations: Bayesian Belief Nets, Markov chains, and Rocchio classification. The algorithms are discussed by many, the variations and types of the RS are usually presented in the handbooks, but what about creating a recommendation system? In his article, Kihn (2015) briefly describes the common types of the RS, and then proceeds to explain how they are created; the recommenders he approaches are the ‘Item-Item Content Recommenders’, ‘Item-Item Collaborative Recommenders’, and ‘User-Item Collaborative Recommenders’.
Kihn (2015) also addresses the Netflix Prize and explains what methods were more or less suitable in that case. A brief history of the RS, prediction algorithms, critical review of the ‘accuracy algorithms’ is presented by Konstan and Riedl (2012). The authors also examine an interesting topic ‘user-recommender lifecycle’ where the process of information filtering, user’s lifecycle, and the connection between benefits to users and community are presented. An important part of the article is ‘Hidden Dangers’ where Konstan and Riedl (2012) explore the risks to users’ privacy and other possible violations.