Browse Books

Reviewer: Maulik A Dave

Internet users typically search for items such as books and computers. Recommender systems provide not only the result of the search, but also a list of other items that the user may be interested in. This book describes many approaches to building recommender systems, ranging from a simple neighborhood approach to complex knowledge-based approaches. The most modern approaches are also covered. The first part covers the basics of recommender systems, and the second part covers modern challenges facing recommendation systems. After a brief overview of the contents of the book in the first chapter, a detailed description of collaborative recommendation follows in chapter 2. Collaborative approaches use the past behavior of the user community for recommendations. The chapter describes memory-based approaches, including user-based nearest neighbor and item-based nearest neighbor. It discusses model-based approaches, in which the raw data is preprocessed, in detail. The discussion of model-based approaches includes associative rules mining and probabilistic approaches. The chapter ends by describing Slope One predictors. Chapter 3 is on content-based approaches, in which the characteristics of items play a central role in recommendations. The vector space model and text classification methods are described in detail, and decision models are mentioned briefly. Chapter 4 is on knowledge-based approaches, in which the user interacts with the system and receives recommendations based on knowledge incorporated in the system. The knowledge-based approaches are classified as constraint-based and case-based. Both types of approaches are discussed in detail, along with some key algorithms and user interaction considerations. The chapter ends with two practical examples. Combining the various approaches is called hybridization, which is the subject of chapter 5. Various kinds of hybridization designs-monolithic, parallel, and pipelined-are presented. Feature combination hybrids, feature augmentation hybrids, hybrids mixed at the user level, hybrids combining weighted scores, hybrids switching depending on situations, hybrids with a sequenced order of techniques, and meta-level hybrids are all discussed. Chapter 6, on explanations in recommender systems, provides mathematical details of well-founded explanations in constraint-based recommenders. It concludes with a brief explanation of the case-based and collaborative recommendation systems. Chapter 7 focuses on the evaluation of recommender systems. It begins by describing general properties of evaluations. The methodology of evaluations based on historical datasets is described later. The chapter ends by showing some experimental designs for evaluation. Chapter 8 explains the evaluation of an example: personalized game recommendations on the Internet. Measurements, described with their results, are "my recommendation," post-sale recommendations, start page recommendations, and overall effects. Chapter 9 is on possible attacks on collaborative recommender systems. It presents various types of attacks, including random, average, bandwagon, segment, nuke, and clickstream attacks. Techniques such as increasing injection costs, additional information on profiles, and automated attack detection are discussed as countermeasures. The chapter ends with a discussion on the privacy aspects of collaborative filtering. Chapter 10 is on aspects of consumer decision making. The factors likely to affect the decision-making processes are discussed, as are personality-based and social psychology-based factors. Chapter 11 is on the challenges facing recommenders in the next generation. Recommenders using explicit trust networks are discussed, followed by a detailed discussion on tag-based recommendations. The chapter concludes by discussing ontological filtering approaches. Chapter 12 covers recommenders in ubiquitous environments. A brief overview of recommendations for various application domains, such as mobile commerce and tourism guides, is presented. Chapter 13 is a summary, and is followed by a 25-page bibliography. For learning about the basics of recommender systems, this book is sufficient, although knowledge of elementary mathematics is necessary to understand the formulas presented. Online Computing Reviews Service