Novelty and diversity metrics for recommender systems: Choice, discovery and relevance
Pablo Castells, Saúl Vargas, Jun Wang · Biblos-e Archivo (Universidad Autónoma de Madrid) · 2011
Abstract. There is an increasing realization in the Recommender Systems (RS) field that novelty and diversity are fundamental qualities of recommendation effectiveness and added-value. We identify however a gap in the formalization of novelty and diversity metrics –and a consensus around them – comparable to the recent proposals in IR diversity. We study a formal characterization of different angles that RS novelty and diversity may take from the end-user viewpoint, aiming to contribute to a formal definition and understanding of different views and meanings of these magnitudes under common groundings. Building upon this, we derive metric schemes that take item position and relevance into account, two aspects not generally addressed in the novelty and diversity metrics reported in the RS literature.