Recommender Systems and Diversity: Taking Advantage of the Long Tail and the Diversity of Recommendation Lists

Version anglaise du chapitre "Recommandeurs et diversité : exploitation de la longue traîne et diversité des listes de recommandations" - ISTEInternational audienceThis chapter presents the stakes linked to the diversity of resources recommended by recommender systems. It describes approaches for evaluating and increasing the diversity within current recommender systems. Recommender systems adapt a selection (filtering) or a resource (adaptation) to a person (personalization), a group of people (group personalization) or a context (e.g. the weather or the user location). Diversity encompasses two major aspects, individual diversity and aggregate diversity. Algorithms implemented by recommender systems are classified into two main categories. Content-based recommender systems typically use information about resources to be recommended, whereas collaborative filtering takes advantage of the set of users via usage data or judgments on resources, be they implicit (consumption, clicks) or explicit (comments, ratings). Various research efforts aim to evaluate the impact of algorithms on diversity, define evaluation metrics and increase individual and aggregate diversity in recommender system