The effect of recommender systems on sales diversity

1 Introduction Recommender systems offer benefits to both consumers and firms. For consumers, recommender systems help individuals both become aware of new products as well as select desirable products among myriad choices (Pham & Healey, 2005). For firms, recommender systems have the potential to increase profits by converting browsers into buyers, cross-selling products, and increasing loyalty through a customized browsing experience (Schafer et al., 1999). Major online firms, such as Amazon.com, Netflix, and Yahoo! Music, host recommender systems, and the prevalence of such systems is expected to grow. In this paper, we study the effect of recommender systems on consumer purchasing of hedonic goods. Two different views exist on the impact of recommenders on diversity of sales. Some believe recommenders will help consumers discover new products (Brynjolffson et al. 2006). On the other hand, others believe the design of common recommender systems will only reinforce the popularity of already popular titles (Mooney & Roy, 1999). The first belief leads to a hypothesis that recommender systems will increase sales diversity. The second belief leads to a hypothesis that recommender systems will decrease sales diversity. This paper is a first attempt to reconcile these seemingly incompatible views. We develop an analytical model of recommender systems and their influence on consumer choice. The model allows us to explore the strong, path-dependent interaction between recommendations and sales. We believe this paper is the first to attempt to isolate the impact of recommender systems on long-run sales diversity. 2 Prior work Despite the growing use of recommender systems, little is known about how these systems affect consumer behavior. In marketing, Cooke et al. (2002) examine how the likelihood of purchasing a recommended product depends on the information provided, context, and familiarity with other products simultaneously recommended. Senecal et al. (2004) experimentally show that recommendations do influence choice. They find that online recommender systems can have even more influence than human recommendations, and the effect is often greater for experience goods than search goods. To the best of our best knowledge, there have been no studies that have attempted to isolate the specific impact of recommenders on sales diversity. However, the topic has received mention from several researchers. For example, Mooney and Roy (1999) suggest that collaborative filters may perpetuate homogeneity in a library setting, but this brief comment is used more to suggest difficulties of working with sparse data than to examine social implications. Brynjolffson …