Recommendation strategies in personalization applications

While the initial goal of recommender systems (RSes) was to reduce the information overload for Internet users and make the information retrieval more efficient, they have become a crucial strategic tool for companies in the online markets. According to this evolution, research on RSes has produced a wide variety of approaches and algorithms. As a consequence, the companies deploying RSes in their business applications face the decision of how to generate and deliver personalized recommendations to their users by choosing among many options. The problem has been largely treated from the machine learning performance perspective because there is relatively little research done from the business perspective. The decision of what kind of recommender engines should be used in a personalization application, given certain business conditions, has a strategic value because it affects the way customers perceive the company with respect to its competitors. Choosing the wrong way to personalize recommendations may not only require the redesign of the information systems but also to rebuild the relationships with customers and even the entire brand strategic positioning. The research issues addressed by this paper are (i) which recommendation strategies a company can deploy to generate and deliver recommendations to users, and (ii) which specific strategies should be used depending on the current business conditions. We propose taxonomy based on a literature analysis and a framework to associate each strategy with a certain setting. The proposed framework is empirically supported by four case studies.