Evaluating recommender systems for supportive technologies

Recommender systems have evolved in recent years into sophisticated support tools that assist users in dealing with the decisions faced in everyday life. Recommender systems were designed to be invaluable in situations, where a large number of options are available, such as deciding what to watch on television, what information to access online, what to purchase in a supermarket, or what to eat. Recommender system evaluations are carried out typically during the design phase of recommender systems to understand the suitability of approaches to the recommendation process, in the usability phase to gain insight into interfacing and user acceptance, and in live user studies to judge the uptake of recommendations generated and impact of the recommender system. In this chapter, we present a detailed overview of evaluation techniques for recommender systems covering a variety of tried and tested methods and metrics. We illustrate their use by presenting a case study that investigates the applicability of a suite of recommender algorithms in a recipe recommender system aimed to assist individuals in planning their daily food intake. The study details an offline evaluation, which compares algorithms, such as collaborative, content-based, and hybrid methods, using multiple performance metrics, to determine the best candidate algorithm for a recipe recommender application.