ABSTRACT
Recommender systems are software tools used by firms to recommend items to users. The list of recommended items is personalized for each user taking into account the user’s preferences. Recommender systems have become increasingly popular over the past decade given their potential to benefit both firms and users and are widely used in a number of industries. Entertainment firms use them for recommending movies (for example, Netflix), music, and videos (for example, YouTube) to consumers. Content websites use such systems to create personalized newspapers, to suggest webpages and documents, and to customize emails containing links to content of various types. Ecommerce firms (such as https://Amazon.com) and service firms use recommendation systems to suggest products, house rentals, matches, and travel-related services. Over the past 15 years, many researchers in marketing, human computer interaction, machine learning, and information retrieval have studied different facets of recommender systems. Adomavicius and Tuzhilin (2005b) provide a comprehensive survey of the early literature on recommender systems. Since this publication, there has been an explosion of research spurred by the Netflix competition. Su and Koshgoftaar (2009) survey many different methods, with a special emphasis on collaborative filtering techniques. The recent monograph by Ekstrand et al. (2010) also focusses on collaborative filtering. The recommender systems handbook, edited by Ricci et al. (2011), contains detailed discussions of different aspects of recommender systems, such as techniques and algorithms, applications, evaluation of recommender systems, and user interactions with recommender systems. Most of the above reviews predominantly concentrate on research done in computer science and allied fields. Given these comprehensive expositions, this chapter will emphasize the contributions from marketing scholars and primarily focus on describing model-based recommender systems. Instead of a detailed review, we will develop a general latent variable framework for user modeling and show how such a framework can subsume a number of approaches that have been explored in the literature. Recommendation systems enable firms to target customers or users on a one-on-one basis. They therefore help ecommerce sites improve product sales and enable content providers to improve the usage of products and services. Firms are primarily interested in increasing the number of items they sell. In addition, firms may be motivated to leverage the heterogeneity of preferences within the customer base to increase the diversity of items that are sold or accessed by users. The hope from the firm’s side is that personalized recommendations will improve customer satisfaction and thereby enhance usage. Users benefit from recommender systems in identifying relevant and preferred items while engaging in less effortful decision making and with reduced information overload. Users may be unaware of the existence of specific alternatives, or may need assistance in sifting through alternatives, particularly in situations that involve too many items. When information changes rapidly and new items are introduced on a regular basis, recommender systems help users by providing relevant information in a timely manner. In some situations, users seek the best sequence of items (such as a television series), or a set or bundle of products that fit well together. In contrast, some users may not be interested in the recommendations per se, but may instead interact with recommendation systems to help others, or to express one’s views.
